docs: add local uv copy-paste install for Antigravity; bump to 0.2.1

README'ye Antigravity için lokal uvx kurulumunu otomatik yapan
kopyala-yapıştır komutu eklendi (~/.gemini/config/mcp_config.json).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

.dockerignore

@@ -181,6 +181,3 @@ site
 
 # Production logs
 **/logs/*.log.*
-**/Dockerfile
-**/Dockerfile
-fly.toml

.env.example

@@ -70,6 +70,50 @@ JWT_SECRET_KEY=your_jwt_secret_key_here
 # MAX_REQUESTS_PER_MINUTE=60
 # BURST_CAPACITY=20
 
+# =============================================================================
+# SEMANTIC SEARCH SETTINGS (Optional)
+# =============================================================================
+
+# Embedding provider for the semantic_search tool.
+# Pick exactly one of: OpenRouter (hosted) or Local (your own server).
+
+# --- Option A: OpenRouter (hosted, default) -----------------------------------
+# Get your API key from: https://openrouter.ai/keys
+# If neither this nor EMBEDDING_PROVIDER=local is set, semantic search is off.
+OPENROUTER_API_KEY=sk-or-v1-your_openrouter_api_key_here
+
+# Optional: override the OpenRouter embedding model and dimension.
+# Defaults: google/gemini-embedding-001 at 3072 dims (paid on OpenRouter).
+# Pick any model from https://openrouter.ai/models?modality=embedding
+# and set the dimension to that model's output size — they must match.
+# OPENROUTER_EMBEDDING_MODEL=google/gemini-embedding-001
+# OPENROUTER_EMBEDDING_DIMENSION=3072
+
+# --- Option B: Local OpenAI-compatible server (no API key required) ----------
+# Recommended for Turkish: intfloat/multilingual-e5-large served by HuggingFace
+# Text Embeddings Inference (TEI). One-line setup:
+#
+#   docker run -p 8080:80 ghcr.io/huggingface/text-embeddings-inference:latest \
+#       --model-id intfloat/multilingual-e5-large
+#
+# Then uncomment the block below. Other model families work too — set
+# EMBEDDING_PROMPT_STYLE to match: e5 / gemini / raw.
+#
+# EMBEDDING_PROVIDER=local
+# LOCAL_EMBEDDING_BASE_URL=http://localhost:8080/v1
+# LOCAL_EMBEDDING_MODEL=intfloat/multilingual-e5-large
+# LOCAL_EMBEDDING_DIMENSION=1024
+# EMBEDDING_PROMPT_STYLE=e5
+# LOCAL_EMBEDDING_API_KEY=                          # most local servers ignore this
+#
+# Ollama fallback (if you prefer Ollama and don't need top Turkish quality):
+#   ollama serve && ollama pull nomic-embed-text
+#   EMBEDDING_PROVIDER=local
+#   LOCAL_EMBEDDING_BASE_URL=http://localhost:11434/v1
+#   LOCAL_EMBEDDING_MODEL=nomic-embed-text
+#   LOCAL_EMBEDDING_DIMENSION=768
+#   EMBEDDING_PROMPT_STYLE=raw   # nomic uses its own search_query/search_document
+
 # =============================================================================
 # USAGE INSTRUCTIONS
 # =============================================================================

.gitignore

@@ -1,3 +1,6 @@
+# Serena
+.serena/
+
 # Byte-compiled / optimized / DLL files
 __pycache__/
 *.py[cod]
@@ -213,3 +216,4 @@ measure_mcp_directly.py
 playwright_mcp_overhead.json
 simple_test.py
 analyze_anayasa_html.py
+CLAUDE.md

CLAUDE.md

@@ -72,7 +72,10 @@ uv run simple_test.py                   # Simple individual tool test
 uv run measure_mcp_directly.py          # Measure MCP token overhead
 
 # Test KVKK module (uses fallback token if BRAVE_API_TOKEN not set)
-python test_kvkk_module.py              # Test KVKK search and document retrieval
+python test_kvkv_module.py              # Test KVKK search and document retrieval
+
+# Test KİK v2 comprehensive functionality
+uv run test_kik_v2_comprehensive.py     # Test all three KİK v2 decision types (uyusmazlik, duzenleyici, mahkeme)
 
 # MCP Optimization Testing
 uv run test_core_tools_quick.py         # Verify all tools work after optimization
@@ -182,7 +185,7 @@ This MCP server has undergone comprehensive optimization to minimize token overh
 4. **emsal_mcp_module**: Emsal (UYAP precedent) decisions
 5. **uyusmazlik_mcp_module**: Uyuşmazlık Mahkemesi (Jurisdictional Disputes Court)
 6. **anayasa_mcp_module**: Constitutional Court (both norm control and individual applications)
-7. **kik_mcp_module**: KİK (Public Procurement Authority) decisions
+7. **kik_mcp_module**: KİK (Public Procurement Authority) decisions with v2 API support for three decision types (uyusmazlik, duzenleyici, mahkeme)
 8. **rekabet_mcp_module**: Competition Authority decisions
 9. **kvkk_mcp_module**: KVKK (Personal Data Protection Authority) decisions - Brave API integration
 10. **bddk_mcp_module**: BDDK (Banking Regulation and Supervision Agency) decisions - Tavily API integration
@@ -466,6 +469,28 @@ doc8 = await get_kvkk_document_markdown(decision_url="https://www.kvkk.gov.tr/Ic
   - **Fallback Token**: If not set, uses a limited free token automatically
   - KVKK search tools will work without configuration (with rate limits)
 
+### Rate Limits
+
+| API | Rate Limit | Notes |
+|-----|------------|-------|
+| Bedesten Unified | ~10 req / 30s window per source IP (measured 2026-05-08); 11th req → HTTP 429 with `Retry-After: 30`. Client uses an internal token bucket (default 1 token, refill 1/3.5s) plus 429 back-pressure (whole bucket pauses for the Retry-After window). Override via `BEDESTEN_RATE_CAPACITY` / `BEDESTEN_RATE_REFILL_S`. |
+| Yargıtay Primary | Unknown | Official government API |
+| Danıştay | Unknown | Official government API |
+| Anayasa Mahkemesi | Unknown | Constitutional Court API |
+| KİK v2 | Unknown | Public Procurement Authority API |
+| Rekabet Kurumu | Unknown | Competition Authority API |
+| Sayıştay | Unknown | Court of Accounts API |
+| Uyuşmazlık | Unknown | Jurisdictional Disputes Court API |
+| Emsal | Unknown | UYAP Precedent Database API |
+| KVKK (Brave) | 1,000/month | Brave Search API free tier limit |
+| BDDK | Unknown | Banking Regulation API |
+
+**Recommendations:**
+- Implement client-side caching for repeated queries
+- Use pagination parameters to limit result sizes
+- Space out requests during bulk operations
+- Consider implementing retry logic with exponential backoff
+
 ### OAuth Authentication Configuration
 
 The server uses **Clerk JWT tokens** for all authentication. **Cross-origin authentication** is implemented using Bearer JWT tokens as per Clerk's best practices.
@@ -1077,15 +1102,21 @@ yargi-mcp
 
 
 ### Date Filtering Format (Bedesten API)
-All Bedesten API tools support consistent date filtering:
-- **Format**: ISO 8601 with Z timezone: `YYYY-MM-DDTHH:MM:SS.000Z`
+All Bedesten API tools support consistent date filtering with **automatic format conversion**:
+- **Accepted Formats**: 
+  - Simple format: `YYYY-MM-DD` (e.g., `2020-01-01`) - **automatically converted**
+  - Full ISO 8601: `YYYY-MM-DDTHH:MM:SS.000Z` (e.g., `2020-01-01T00:00:00.000Z`)
+- **Automatic Conversion**: ✅ **NEW FEATURE** - Simple dates are automatically converted to ISO 8601 format
+  - Start dates: `2020-01-01` → `2020-01-01T00:00:00.000Z`
+  - End dates: `2020-01-01` → `2020-01-01T23:59:59.999Z`
 - **Parameters**: `kararTarihiStart` (start date) and `kararTarihiEnd` (end date)
 - **Usage**: Both parameters are optional, use together for date ranges or single parameter for one-sided filtering
 - **Examples**:
-  - Single date: `kararTarihiStart="2024-06-25T00:00:00.000Z", kararTarihiEnd="2024-06-25T23:59:59.999Z"`
-  - Year range: `kararTarihiStart="2024-01-01T00:00:00.000Z", kararTarihiEnd="2024-12-31T23:59:59.999Z"`
-  - From date: `kararTarihiStart="2024-01-01T00:00:00.000Z"` (no end date)
-  - Until date: `kararTarihiEnd="2024-12-31T23:59:59.999Z"` (no start date)
+  - **Simple format**: `kararTarihiStart="2024-06-25", kararTarihiEnd="2024-06-25"` ✅ **Works automatically**
+  - **Year range**: `kararTarihiStart="2024-01-01", kararTarihiEnd="2024-12-31"` ✅ **Auto-converted**
+  - **Full ISO format**: `kararTarihiStart="2024-01-01T00:00:00.000Z", kararTarihiEnd="2024-12-31T23:59:59.999Z"`
+  - **From date**: `kararTarihiStart="2024-01-01"` (no end date)
+  - **Until date**: `kararTarihiEnd="2024-12-31"` (no start date)
 
 ### Exact Phrase Search Format (Bedesten API)
 All Bedesten API tools support two types of phrase searching:
@@ -1542,13 +1573,25 @@ This ASGI support transforms the Yargı MCP server into a versatile web service
 **Authentication**: ✅ **OAuth 2.0 + Bearer JWT** - Cross-origin authentication working
 **Last Updated**: 2025-01-21 - All critical issues resolved
 
+**Free Deployment**: `https://yargi-mcp-free.fly.dev` - No authentication required
+**Status**: ✅ **OPERATIONAL** - Authorization disabled for open access
+**Use Case**: Development, testing, and open-source usage without OAuth setup
+
 #### Live Production Endpoints
+
+**Authenticated Deployment (api.yargimcp.com)**:
 - **Health Check**: https://api.yargimcp.com/health
 - **OAuth Login**: https://api.yargimcp.com/auth/login  
 - **MCP Endpoint (HTTP)**: https://api.yargimcp.com/mcp/
 - **MCP Endpoint (SSE)**: https://api.yargimcp.com/sse/
 - **OAuth Discovery**: https://api.yargimcp.com/.well-known/oauth-authorization-server
 
+**Free Deployment (yargi-mcp-free.fly.dev)**:
+- **Health Check**: https://yargi-mcp-free.fly.dev/health
+- **MCP Endpoint (HTTP)**: https://yargi-mcp-free.fly.dev/mcp/
+- **MCP Endpoint (SSE)**: https://yargi-mcp-free.fly.dev/sse/
+- **Direct Access**: No authentication required - immediate usage
+
 ### Redis Configuration (Fly.io Native Upstash) ✅
 
 The server uses Fly.io's native Upstash Redis integration for OAuth session storage:
@@ -1631,6 +1674,8 @@ ENABLE_AUTH=true
 ```
 
 #### MCP Connection Details for Claude AI
+
+**Authenticated Production (api.yargimcp.com)**:
 ```
 MCP Server URL (HTTP): https://api.yargimcp.com/mcp/
 MCP Server URL (SSE): https://api.yargimcp.com/sse/
@@ -1640,6 +1685,15 @@ Authentication: OAuth 2.0 with PKCE + JWT tokens + Bearer JWT (optional)
 Transports: HTTP (Streamable) + SSE (Server-Sent Events)
 ```
 
+**Free Open Access (yargi-mcp-free.fly.dev)**:
+```
+MCP Server URL (HTTP): https://yargi-mcp-free.fly.dev/mcp/
+MCP Server URL (SSE): https://yargi-mcp-free.fly.dev/sse/
+Authentication: None - Direct access
+Transports: HTTP (Streamable) + SSE (Server-Sent Events)
+Use Case: Development, testing, immediate usage without OAuth setup
+```
+
 #### SSE Transport Implementation ✅
 
 The server now supports **Server-Sent Events (SSE)** transport alongside HTTP:
@@ -2031,7 +2085,7 @@ build-backend = "setuptools.build_meta"
 4. **Emsal**: 2 tools (search + document)
 5. **Uyuşmazlık**: 2 tools (search + document)
 6. **Constitutional Court**: ✅ 2 tools (unified norm control + individual applications) - **NEWLY UNIFIED**
-7. **KİK**: 2 tools (search + document)
+7. **KİK**: 2 tools (search + document) - **v2 API with three decision types: uyusmazlik, duzenleyici, mahkeme** ✅
 8. **Competition Authority**: 2 tools (search + document)
 9. **KVKK**: 2 tools (search + document)
 10. **Sayıştay**: 4 tools (3 search types + document)
@@ -2069,6 +2123,52 @@ build-backend = "setuptools.build_meta"
   - ✅ Session management and tool discovery
 - **Claude AI**: Successfully connects and uses all Turkish legal database tools
 
+#### ✅ Bedesten Tools Null Safety Fixes (Completed - Jul 23, 2025)
+- **Issue**: TypeError "cannot convert undefined or null to object" in Bedesten search and document tools
+- **Root Cause**: API responses containing null/undefined fields without proper validation
+- **Fixes Applied**:
+  - ✅ **Search Function**: Added null safety checks for `response.data.emsalKararList` and `response.data.total`
+  - ✅ **Document Function**: Added comprehensive validation for `doc_response.data`, `content`, and `mimeType` fields
+  - ✅ **Error Handling**: Added descriptive error messages and graceful fallbacks
+  - ✅ **Base64 Decoding**: Protected base64 operations with try-catch blocks
+- **Result**: Bedesten tools now handle API edge cases gracefully without crashing
+- **Production Status**: Deployed and operational on api.yargimcp.com
+
+#### ✅ Automatic Date Format Conversion (Completed - Sep 2, 2025)
+- **Issue**: Bedesten API requires ISO 8601 format with timezone, but users were providing simple dates
+- **Problem**: Queries like `kararTarihiStart="2020-01-01"` returned "No data returned from Bedesten API"
+- **Root Cause**: Simple date format `YYYY-MM-DD` not converted to required `YYYY-MM-DDTHH:MM:SS.000Z` format
+- **Solution Applied**: ✅ **Automatic Date Format Conversion** in `search_bedesten_unified`
+  - **Start dates**: `2020-01-01` → `2020-01-01T00:00:00.000Z` (beginning of day)
+  - **End dates**: `2020-01-01` → `2020-01-01T23:59:59.999Z` (end of day)
+  - **Backwards compatible**: Full ISO 8601 dates still work unchanged
+  - **Smart detection**: Only converts if date doesn't already end with 'Z'
+- **Benefits**: 
+  - ✅ **User-friendly**: Simple date input now works seamlessly
+  - ✅ **Inclusive ranges**: End dates include the entire specified day
+  - ✅ **No breaking changes**: Existing ISO 8601 usage unaffected
+- **Production Status**: Deployed on api.yargimcp.com - **Version 353**
+
+#### ✅ KİK v2 MCP Implementation Testing (Completed - Sep 2, 2025)
+- **Issue**: Test KİK v2 MCP implementation with all three decision types
+- **Request**: "üç karar türü ile de mcpyi test et" (test the MCP with all three decision types)  
+- **Decision Types Tested**:
+  - ✅ **uyusmazlik** (dispute) - 500 decisions found and searchable
+  - ✅ **duzenleyici** (regulatory) - 8 decisions found and searchable  
+  - ✅ **mahkeme** (court) - 318 decisions found and searchable
+- **Total Coverage**: 826 decisions across all three decision types
+- **SSL Issues**: ✅ Resolved with legacy server connect configuration  
+- **API Endpoints**: All three endpoints working correctly
+  - `/api/KurulKararlari/GetKurulKararlari` (uyusmazlik)
+  - `/api/KurulKararlari/GetKurulKararlariDk` (duzenleyici)  
+  - `/api/KurulKararlari/GetKurulKararlariMk` (mahkeme)
+- **Hash Analysis**: Comprehensive testing performed to understand document ID encryption
+  - Tested various hash generation patterns (SHA256, HMAC, composite hashes)
+  - Angular/cryptoService.encrypt() style approaches tested
+  - Hash eşleşmesi bulunamadı - client-side session data veya farklı algoritma kullanılıyor olabilir
+- **Result**: ✅ KİK v2 MCP implementation fully operational for all three decision types
+- **Production Status**: All tests passing, search functionality working, ready for production deployment
+
 ### Key Features
 - **FastMCP Framework**: Modern MCP server implementation
 - **Unified APIs**: Single interface for multiple court systems

Dockerfile

@@ -1,29 +1,53 @@
-# -------- BASE IMAGE (includes Chromium & deps) ----------------------------
-FROM mcr.microsoft.com/playwright/python:v1.53.0-noble
+# Use Python 3.12 slim image
+FROM python:3.12-slim
 
-# -------- Runtime setup ----------------------------------------------------
+# Set working directory
 WORKDIR /app
 
-# Copy dependency manifests first for layer-cache
-COPY pyproject.toml poetry.lock* requirements*.txt* ./
+# Install system dependencies (gcc/g++ kept in case any wheel falls back to source build)
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    gcc \
+    g++ \
+    && rm -rf /var/lib/apt/lists/*
 
-# Fast, deterministic install with `uv`
-RUN pip install --no-cache-dir uv && \
-    uv pip install --system --no-cache-dir .[asgi,saas]
+# Copy project metadata first for better Docker layer caching
+COPY pyproject.toml ./
+COPY README.md ./
 
-# Copy application source
-COPY . .
+# Copy entry points
+COPY app.py ./
+COPY asgi_app.py ./
+COPY mcp_server_main.py ./
 
-# -------- Environment ------------------------------------------------------
-ENV PYTHONUNBUFFERED=1
-ENV ENABLE_AUTH=true
-ENV PORT=8000
+# Copy MCP modules and shared packages
+COPY anayasa_mcp_module ./anayasa_mcp_module
+COPY bddk_mcp_module ./bddk_mcp_module
+COPY bedesten_mcp_module ./bedesten_mcp_module
+COPY danistay_mcp_module ./danistay_mcp_module
+COPY emsal_mcp_module ./emsal_mcp_module
+COPY gib_mcp_module ./gib_mcp_module
+COPY kik_mcp_module ./kik_mcp_module
+COPY kvkk_mcp_module ./kvkk_mcp_module
+COPY rekabet_mcp_module ./rekabet_mcp_module
+COPY sayistay_mcp_module ./sayistay_mcp_module
+COPY sigorta_tahkim_mcp_module ./sigorta_tahkim_mcp_module
+COPY uyusmazlik_mcp_module ./uyusmazlik_mcp_module
+COPY yargitay_mcp_module ./yargitay_mcp_module
+COPY semantic_search ./semantic_search
 
-# -------- Health check -----------------------------------------------------
-HEALTHCHECK --interval=30s --timeout=10s --start-period=10s --retries=3 \
-  CMD python -c "import httpx, os, sys; r=httpx.get(f'http://localhost:{os.getenv(\"PORT\",\"8000\")}/health'); sys.exit(0 if r.status_code==200 else 1)"
+# Install the package with ASGI extras (uvicorn + starlette)
+RUN pip install --no-cache-dir -e ".[asgi]"
 
+# Expose port
 EXPOSE 8000
 
-# -------- Entrypoint -------------------------------------------------------
-CMD ["uvicorn", "asgi_app:app", "--host", "0.0.0.0", "--port", "8000", "--proxy-headers"]
+# Set environment variables
+ENV PORT=8000
+ENV PYTHONUNBUFFERED=1
+
+# Health check
+HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \
+    CMD python -c "import httpx; httpx.get('http://localhost:8000/health', timeout=5)" || exit 1
+
+# Run the ASGI application
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8000"]

Procfile

@@ -1 +0,0 @@
-web: uvicorn asgi_app:app --host 0.0.0.0 --port $PORT

README.md

@@ -1,8 +1,103 @@
 # Yargı MCP: Türk Hukuk Kaynakları için MCP Sunucusu
 
+> ## ✨ Profesyonel Sürüm Hazır: Yargı MCP Pro
+>
+> **Mevzuat ve içtihatı tek bir MCP sunucusunda birleştiren** profesyonel sürüm yayında:
+>
+> 👉 **https://yargi.betaspacestudio.com**
+
+> ## 🚨 SUNUCU YENİ ADRESE TAŞINDI
+>
+> **Yeni Remote MCP adresi:** `https://yargimcp.surucu.dev/mcp`
+>
+> **Eski adres** (`https://yargimcp.fastmcp.app/mcp`) **artık kullanım dışıdır** — yalnızca taşındığını bildiren bir uyarı tool'u döner.
+>
+> **Yapmanız gereken:** MCP istemcinizdeki (Claude Desktop, 5ire, Google Antigravity, ChatGPT vb.) sunucu URL'sini yukarıdaki yeni adresle güncelleyin.
+
+## Word'den UDF'ye profesyonel dönüşüm için yeni uygulamam [udfcevir.com](https://udfcevir.com) adresinde! 
+
 [![Star History Chart](https://api.star-history.com/svg?repos=saidsurucu/yargi-mcp&type=Date)](https://www.star-history.com/#saidsurucu/yargi-mcp&Date)
 
-Bu proje, çeşitli Türk hukuk kaynaklarına (Yargıtay, Danıştay, Emsal Kararlar, Uyuşmazlık Mahkemesi, Anayasa Mahkemesi - Norm Denetimi ile Bireysel Başvuru Kararları, Kamu İhale Kurulu Kararları, Rekabet Kurumu Kararları, Sayıştay Kararları, KVKK Kararları ve BDDK Kararları) erişimi kolaylaştıran bir [FastMCP](https://gofastmcp.com/) sunucusu oluşturur. Bu sayede, bu kaynaklardan veri arama ve belge getirme işlemleri, Model Context Protocol (MCP) destekleyen LLM (Büyük Dil Modeli) uygulamaları (örneğin Claude Desktop veya [5ire](https://5ire.app)) ve diğer istemciler tarafından araç (tool) olarak kullanılabilir hale gelir.
+Bu proje, çeşitli Türk hukuk kaynaklarına (Yargıtay, Danıştay, Emsal Kararlar, Uyuşmazlık Mahkemesi, Anayasa Mahkemesi - Norm Denetimi ile Bireysel Başvuru Kararları, Kamu İhale Kurulu Kararları, Rekabet Kurumu Kararları, Sayıştay Kararları, KVKK Kararları, BDDK Kararları, GİB Özelgeleri ve Sigorta Tahkim Komisyonu Kararları) erişimi kolaylaştıran bir [FastMCP](https://gofastmcp.com/) sunucusu oluşturur. Bu sayede, bu kaynaklardan veri arama ve belge getirme işlemleri, Model Context Protocol (MCP) destekleyen LLM (Büyük Dil Modeli) uygulamaları (örneğin Claude Desktop veya [5ire](https://5ire.app)) ve diğer istemciler tarafından araç (tool) olarak kullanılabilir hale gelir.
+
+---
+
+## 🚀 5 Dakikada Başla (Remote MCP)
+
+### ✅ Kurulum Gerektirmez! Hemen Kullan!
+
+🔗 **Remote MCP Adresi:** `https://yargimcp.surucu.dev/mcp`
+
+> ⚠️ **Eski adres** `https://yargimcp.fastmcp.app/mcp` **artık kullanım dışıdır** — yalnızca taşındığını bildiren bir uyarı tool'u döner. Lütfen yukarıdaki yeni adresi kullanın.
+
+### Claude Desktop ile Kullanım (Ücretli abonelik gerekir)
+
+1. **Claude Desktop'ı açın**
+2. **Settings → Connectors → Add Custom Connector**
+3. **Bilgileri girin:**
+   - **Name:** `Yargı MCP`
+   - **URL:** `https://yargimcp.surucu.dev/mcp`
+4. **Add** butonuna tıklayın
+5. **Hemen kullanmaya başlayın!** 🎉
+
+### Google Antigravity ile Kullanım (Lokal `uv` Kurulumu — Kopyala-Yapıştır)
+
+> **Ön Gereksinimler:** Bilgisayarınızda **Python**, **`uv`** ([kurulum](https://docs.astral.sh/uv/getting-started/installation/)) ve **Node.js** ([indir](https://nodejs.org/en/download)) kurulu olmalı. (Node.js yalnızca aşağıdaki kurulum komutunu çalıştırmak için gerekir; MCP'yi `uvx` çalıştırır.)
+
+Aşağıdaki **bloğun tamamını** terminale yapıştırın. Komut, Antigravity'nin okuduğu `~/.gemini/config/mcp_config.json` dosyasını sizin yerinize oluşturur/günceller (varsa diğer sunucularınız korunur):
+
+**macOS / Linux** (Terminal):
+
+```bash
+node - <<'YARGI'
+const fs=require("fs"),os=require("os"),path=require("path");
+const dir=path.join(os.homedir(),".gemini","config"),file=path.join(dir,"mcp_config.json");
+fs.mkdirSync(dir,{recursive:true});
+let cfg={};try{cfg=JSON.parse(fs.readFileSync(file,"utf8"))}catch{}
+if(typeof cfg!=="object"||cfg===null||Array.isArray(cfg))cfg={};
+if(typeof cfg.mcpServers!=="object"||cfg.mcpServers===null)cfg.mcpServers={};
+cfg.mcpServers["yargi-mcp"]={command:"uvx",args:["yargi-mcp"]};
+fs.writeFileSync(file,JSON.stringify(cfg,null,2)+"\n");
+console.log("yargi-mcp eklendi -> "+file);
+YARGI
+```
+
+**Windows** (PowerShell):
+
+```powershell
+@'
+const fs=require("fs"),os=require("os"),path=require("path");
+const dir=path.join(os.homedir(),".gemini","config"),file=path.join(dir,"mcp_config.json");
+fs.mkdirSync(dir,{recursive:true});
+let cfg={};try{cfg=JSON.parse(fs.readFileSync(file,"utf8"))}catch{}
+if(typeof cfg!=="object"||cfg===null||Array.isArray(cfg))cfg={};
+if(typeof cfg.mcpServers!=="object"||cfg.mcpServers===null)cfg.mcpServers={};
+cfg.mcpServers["yargi-mcp"]={command:"uvx",args:["yargi-mcp"]};
+fs.writeFileSync(file,JSON.stringify(cfg,null,2)+"\n");
+console.log("yargi-mcp eklendi -> "+file);
+'@ | node -
+```
+
+Komut `yargi-mcp eklendi -> ...` çıktısını verdiğinde kurulum tamamlanmıştır. Antigravity'yi (açıksa kapatıp) yeniden başlatın; `yargi-mcp` araçları otomatik yüklenir.
+
+> 💡 **İpucu:** Lokal kurulumda hukuk kaynaklarına erişim doğrudan bilgisayarınızda `uvx yargi-mcp` ile çalışır; uzaktan sunucuya ihtiyaç duymaz.
+
+### Remote MCP Sorun Giderme
+
+`https://yargimcp.surucu.dev/mcp` bir web sayfası değil, Streamable HTTP MCP uç noktasıdır. Tarayıcıda açınca veya düz `curl` ile GET isteği atınca `406 Not Acceptable` ve `Client must accept text/event-stream` benzeri bir yanıt görmek normaldir; bu, sunucunun kapalı olduğu anlamına gelmez. MCP istemcisi `Accept: application/json, text/event-stream` başlığıyla JSON-RPC isteği göndermelidir.
+
+Hızlı sağlık kontrolü için tarayıcıda şu adresleri açabilirsiniz:
+
+- `https://yargimcp.surucu.dev/health` — servis sağlık durumu
+
+Claude.ai veya başka bir istemci "araç yok" gibi davranırsa:
+
+1. Connector'ı kaldırıp yeniden ekleyin.
+2. URL olarak önce `https://yargimcp.surucu.dev/mcp` deneyin; istemciniz yönlendirmeleri takip etmiyorsa `https://yargimcp.surucu.dev/mcp/` deneyin.
+3. Eski `https://yargimcp.fastmcp.app/mcp` adresinin istemci ayarlarında veya önbellekte kalmadığından emin olun.
+4. İstemcinin remote/Streamable HTTP MCP desteklediğini ve `text/event-stream` kabul ettiğini kontrol edin.
+
+---
 
 ![örnek](./ornek.png)
 
@@ -30,6 +125,8 @@ Bu proje, çeşitli Türk hukuk kaynaklarına (Yargıtay, Danıştay, Emsal Kara
     * **Sayıştay:** 3 karar türü ile kapsamlı denetim kararlarına erişim + **8 Daire Filtreleme** + **Tarih Aralığı & İçerik Arama** (Genel Kurul yorumlayıcı kararları, Temyiz Kurulu itiraz kararları, Daire ilk derece denetim kararları)
     * **KVKK (Kişisel Verilerin Korunması Kurulu):** Brave Search API ile veri koruma kararlarını arama; uzun karar metinlerini (5.000 karakterlik) sayfalanmış Markdown formatında getirme + **Türkçe Arama** + **Site Hedeflemeli Arama** (kvkk.gov.tr kararları)
     * **BDDK (Bankacılık Düzenleme ve Denetleme Kurumu):** Bankacılık düzenleme kararlarını arama; karar metinlerini Markdown formatında getirme + **Optimized Search** + **"Karar Sayısı" Targeting** + **Spesifik URL Filtreleme** (bddk.org.tr/Mevzuat/DokumanGetir)
+    * **GİB (Gelir İdaresi Başkanlığı) Özelgeleri:** Resmi vergi özelgelerini arama (18.000+ özelge: KDV, Kurumlar, Gelir, ÖTV, Damga vb.); tam metni sayfalanmış Markdown formatında getirme + **Keyword + Özelge No + Kanun No + Tarih Aralığı** + **Otomatik ISO 8601 Dönüşümü** + **Metadata Başlık Bloğu**
+    * **Sigorta Tahkim Komisyonu:** Hakem Karar Dergisi (64 sayı, 2010-2025) içindeki sigorta tahkim kararlarını arama; dergi PDF'lerini Markdown formatında getirme + **Sayı İçi Karar Arama** + **Türkçe Büyük/Küçük Harf Desteği** + **Relevance Scoring**
 
 * Karar metinlerinin daha kolay işlenebilmesi için Markdown formatına çevrilmesi.
 * Claude Desktop uygulaması ile `fastmcp install` komutu kullanılarak kolay entegrasyon.
@@ -132,10 +229,110 @@ Yargı MCP'yi Gemini CLI ile kullanmak için:
 
 </details>
 
+---
+<details>
+<summary>🧠 <strong>Semantik Arama (Opsiyonel)</strong></summary>
+
+Yargı MCP, **semantik arama** özelliği ile kararları anlamsal olarak sıralayabilir. Opsiyoneldir; iki yoldan biri yapılandırıldığında otomatik etkinleşir:
+
+- **Yerel** (önerilen, ücretsiz): kendi makinenizdeki OpenAI-uyumlu embedding sunucusu (HuggingFace TEI, llama.cpp, Ollama, vLLM, LM Studio…)
+- **Hosted**: OpenRouter API anahtarı
+
+### Semantik Arama Nasıl Çalışır?
+1. `initial_keyword` ile Bedesten API'den 100 karar çekilir
+2. `query` ile bu kararlar embedding modeli kullanılarak anlamsal olarak sıralanır
+3. En alakalı kararlar döndürülür
+
+### Önerilen Türkçe Kurulumu (Yerel — `multilingual-e5-large`)
+
+`intfloat/multilingual-e5-large` Türkçe için kıyas ettiğimiz açık kaynak modeller arasında en iyilerinden. HuggingFace'in **Text Embeddings Inference (TEI)** sunucusuyla tek komutta ayağa kalkar ve OpenAI-uyumlu API sunar:
+
+```bash
+docker run -p 8080:80 ghcr.io/huggingface/text-embeddings-inference:latest \
+    --model-id intfloat/multilingual-e5-large
+```
+
+Sonra Yargı MCP'ye şu env vars'ları geçirin:
+
+```bash
+EMBEDDING_PROVIDER=local
+LOCAL_EMBEDDING_BASE_URL=http://localhost:8080/v1
+LOCAL_EMBEDDING_MODEL=intfloat/multilingual-e5-large
+LOCAL_EMBEDDING_DIMENSION=1024
+EMBEDDING_PROMPT_STYLE=e5
+```
+
+> ⚠️ **Önemli:** `EMBEDDING_PROMPT_STYLE=e5` şart — e5 modelleri `query:` / `passage:` öneki bekleyecek şekilde eğitilmiştir; yanlış önek sessizce kaliteyi düşürür.
+
+#### Claude Desktop örneği (yerel TEI)
+```json
+{
+  "mcpServers": {
+    "Yargı MCP": {
+      "command": "uvx",
+      "args": ["yargi-mcp"],
+      "env": {
+        "EMBEDDING_PROVIDER": "local",
+        "LOCAL_EMBEDDING_BASE_URL": "http://localhost:8080/v1",
+        "LOCAL_EMBEDDING_MODEL": "intfloat/multilingual-e5-large",
+        "LOCAL_EMBEDDING_DIMENSION": "1024",
+        "EMBEDDING_PROMPT_STYLE": "e5"
+      }
+    }
+  }
+}
+```
+
+### Alternatif 1: Ollama (yerel, daha hafif kurulum)
+
+```bash
+ollama serve
+ollama pull nomic-embed-text   # 768 dim, İngilizce ağırlıklı
+```
+
+```bash
+EMBEDDING_PROVIDER=local
+LOCAL_EMBEDDING_BASE_URL=http://localhost:11434/v1
+LOCAL_EMBEDDING_MODEL=nomic-embed-text
+LOCAL_EMBEDDING_DIMENSION=768
+EMBEDDING_PROMPT_STYLE=raw
+```
+
+> Ollama kütüphanesinde `multilingual-e5-large` doğrudan yok; Türkçe için TEI yolu daha doğru sonuç verir.
+
+### Alternatif 2: OpenRouter (hosted)
+
+```bash
+OPENROUTER_API_KEY=sk-or-v1-xxx...
+# İsteğe bağlı — varsayılan google/gemini-embedding-001 (3072 dim, ÜCRETLİ)
+# OPENROUTER_EMBEDDING_MODEL=...
+# OPENROUTER_EMBEDDING_DIMENSION=...
+# EMBEDDING_PROMPT_STYLE=gemini   # varsayılan
+```
+
+API anahtarınızı [openrouter.ai/keys](https://openrouter.ai/keys) adresinden alın. Varsayılan model `google/gemini-embedding-001` artık ücretli — ücretsiz bir model seçerseniz `OPENROUTER_EMBEDDING_MODEL`, `OPENROUTER_EMBEDDING_DIMENSION` ve uygun `EMBEDDING_PROMPT_STYLE` değerlerini birlikte ayarlayın.
+
+### Yapılandırma Referansı
+
+| Env Var | Açıklama | Örnek |
+|---|---|---|
+| `EMBEDDING_PROVIDER` | `local` ise yerel sunucu, boş ise OpenRouter | `local` |
+| `EMBEDDING_PROMPT_STYLE` | `gemini` / `e5` / `raw` — modelin beklediği önek | `e5` |
+| `LOCAL_EMBEDDING_BASE_URL` | Yerel sunucunun OpenAI-uyumlu URL'i | `http://localhost:8080/v1` |
+| `LOCAL_EMBEDDING_MODEL` | Model adı | `intfloat/multilingual-e5-large` |
+| `LOCAL_EMBEDDING_DIMENSION` | Modelin çıktı boyutu (mutlaka eşleşmeli) | `1024` |
+| `OPENROUTER_API_KEY` | OpenRouter anahtarı (sadece hosted için) | `sk-or-v1-…` |
+| `OPENROUTER_EMBEDDING_MODEL` | OpenRouter model id'si | `google/gemini-embedding-001` |
+| `OPENROUTER_EMBEDDING_DIMENSION` | OpenRouter modelinin çıktı boyutu | `3072` |
+
+> 💡 **Not:** Hiçbir embedding sağlayıcı yapılandırılmazsa semantik arama aracı görünmez, diğer 24 araç normal şekilde çalışır.
+
+</details>
+
 <details>
 <summary>🛠️ <strong>Kullanılabilir Araçlar (MCP Tools)</strong></summary>
 
-Bu FastMCP sunucusu **19 optimize edilmiş MCP aracı** sunar (token verimliliği için optimize edilmiş):
+Bu FastMCP sunucusu **26 aktif MCP aracı** + **1 opsiyonel semantik arama aracı** sunar (token verimliliği için optimize edilmiş):
 
 ### **Yargıtay Araçları (Birleşik Bedesten API - Token Optimized)**
 *Not: Yargıtay araçları token verimliliği için birleşik Bedesten API'ye entegre edilmiştir*
@@ -160,8 +357,8 @@ Bu FastMCP sunucusu **19 optimize edilmiş MCP aracı** sunar (token verimliliğ
 8. `get_anayasa_document_unified(document_url, page_number)`: AYM kararlarını birleşik belge getirme - **sayfalanmış Markdown** içeriği
 
 ### **KİK (Kamu İhale Kurulu) Araçları**
-9. `search_kik_decisions(karar_tipi, ...)`: KİK (Kamu İhale Kurulu) kararlarını arar. 
-10. `get_kik_document_markdown(karar_id, page_number)`: Belirli bir KİK kararını, Base64 ile encode edilmiş `karar_id`'sini kullanarak alır ve **sayfalanmış Markdown** içeriğini getirir.
+9. `search_kik_v2_decisions(decision_type, karar_metni, karar_no, basvuran, idare_adi, baslangic_tarihi, bitis_tarihi)`: KİK v2 API ile uyuşmazlık, düzenleyici ve mahkeme kararlarını arar.
+10. `get_kik_v2_document_markdown(gundemMaddesiId)`: Arama sonucundaki `gundemMaddesiId` ile KİK karar metnini Markdown formatında getirir.
 ### **Rekabet Kurumu Araçları**
     * `search_rekabet_kurumu_decisions(KararTuru: Literal[...], ...) -> RekabetSearchResult`: Rekabet Kurumu kararlarını arar. `KararTuru` için kullanıcı dostu isimler kullanılır (örn: "Birleşme ve Devralma").
     * `get_rekabet_kurumu_document(karar_id: str, page_number: Optional[int] = 1) -> RekabetDocument`: Belirli bir Rekabet Kurumu kararını `karar_id` ile alır. Kararın PDF formatındaki orijinalinden istenen sayfayı ayıklar ve Markdown formatında döndürür.
@@ -169,22 +366,32 @@ Bu FastMCP sunucusu **19 optimize edilmiş MCP aracı** sunar (token verimliliğ
 
 ---
 
-* **Sayıştay Araçları (3 Karar Türü + 8 Daire Filtreleme):**
-    * `search_sayistay_genel_kurul(karar_no, karar_tarih_baslangic, karar_tamami, ...)`: Sayıştay Genel Kurul (yorumlayıcı) kararlarını arar. **Tarih aralığı** (2006-2024) + **İçerik arama** (400 karakter)
-    * `search_sayistay_temyiz_kurulu(ilam_dairesi, kamu_idaresi_turu, temyiz_karar, ...)`: Temyiz Kurulu (itiraz) kararlarını arar. **8 Daire filtreleme** + **Kurum türü** + **Konu sınıflandırması**
-    * `search_sayistay_daire(yargilama_dairesi, web_karar_metni, hesap_yili, ...)`: Daire (ilk derece denetim) kararlarını arar. **8 Daire filtreleme** + **Hesap yılı** + **İçerik arama**
-    * `get_sayistay_genel_kurul_document_markdown(decision_id: str)`: Genel Kurul kararının tam metnini Markdown formatında getirir
-    * `get_sayistay_temyiz_kurulu_document_markdown(decision_id: str)`: Temyiz Kurulu kararının tam metnini Markdown formatında getirir  
-    * `get_sayistay_daire_document_markdown(decision_id: str)`: Daire kararının tam metnini Markdown formatında getirir
+* **Sayıştay Araçları (Birleşik API, 3 Karar Türü + 8 Daire Filtreleme):**
+    * `search_sayistay_unified(decision_type, start, length, ...)`: `genel_kurul`, `temyiz_kurulu` veya `daire` kararlarını tek araçla arar. `length` 1-100 aralığındadır.
+    * `get_sayistay_document_unified(decision_id, decision_type)`: Birleşik arama sonucundaki karar ID'si ve karar türüyle tam metni Markdown formatında getirir.
 
 * **KVKK Araçları (Brave Search API + Türkçe Arama):**
-    * `search_kvkk_decisions(keywords, page, pageSize, ...)`: KVKK (Kişisel Verilerin Korunması Kurulu) kararlarını Brave Search API ile arar. **Türkçe arama** + **Site hedeflemeli** (`site:kvkk.gov.tr "karar özeti"`) + **Sayfalama desteği**
+    * `search_kvkk_decisions(keywords, page)`: KVKK (Kişisel Verilerin Korunması Kurulu) kararlarını Brave Search API ile arar. **Türkçe arama** + **Site hedeflemeli** (`site:kvkk.gov.tr "karar özeti"`) + **Sayfalama desteği**. Sonuç sayısı sunucuda 10 olarak sabitlenmiştir.
     * `get_kvkk_document_markdown(decision_url: str, page_number: Optional[int] = 1)`: KVKK kararının tam metnini **sayfalanmış Markdown** formatında getirir (5.000 karakterlik sayfa)
 
 ### BDDK Araçları
     * `search_bddk_decisions(keywords, page)`: BDDK (Bankacılık Düzenleme ve Denetleme Kurumu) kararlarını arar. **"Karar Sayısı" targeting** + **Spesifik URL filtreleme** (`bddk.org.tr/Mevzuat/DokumanGetir`) + **Optimized search**
     * `get_bddk_document_markdown(document_id: str, page_number: Optional[int] = 1)`: BDDK kararının tam metnini **sayfalanmış Markdown** formatında getirir (5.000 karakterlik sayfa)
 
+### GİB (Gelir İdaresi Başkanlığı) Özelge Araçları (Resmi GİB JSON API)
+    * `search_gib_ozelge(keywords, ozelgeNo, kanunNo, ozelgeStartDate, ozelgeEndDate, page, pageSize)`: GİB özelgelerini (Türk Gelir İdaresi Başkanlığı vergi özelgeleri) arar — **18.000+ özelge** (KDV, Kurumlar, Gelir, ÖTV, Damga, VUK vb.). **Keyword + Özelge No + Kanun No + Tarih Aralığı** + **Otomatik ISO 8601 Dönüşümü** (`YYYY-MM-DD` girdileri otomatik olarak full ISO 8601'e çevrilir)
+    * `get_gib_ozelge_document_markdown(ozelge_id: int, page_number: int = 1)`: Belirli bir özelgenin tam metnini **sayfalanmış Markdown** formatında getirir (5.000 karakterlik sayfa) + **Metadata başlık bloğu** (Başlık, Sayı, Tarih, Kanun, Kaynak URL)
+
+### Sigorta Tahkim Komisyonu Araçları (Tavily Search API + PDF)
+    * `search_sigorta_tahkim_decisions(keywords, page)`: Sigorta Tahkim Komisyonu kararlarını Tavily Search API ile arar. **Site hedeflemeli** (`sigortatahkim.org`) + **Sayfalama desteği**. Sonuç sayısı sunucuda 10 olarak sabitlenmiştir.
+    * `get_sigorta_tahkim_document_markdown(issue_number: str, page_number: int)`: Hakem Karar Dergisi sayısının PDF'ini indirip **sayfalanmış Markdown** formatında getirir (5.000 karakterlik sayfa). 64 sayı (2010-2025)
+    * `search_within_sigorta_tahkim_issue(issue_number: str, keyword: str, max_results: int)`: Belirli bir dergi sayısı içindeki kararları anahtar kelime ile arar. **Türkçe İ/I desteği** + **Relevance scoring** + **Excerpt** ile sonuç
+
+### Yardımcı ve Uyumluluk Araçları
+    * `check_government_servers_health()`: Yargı kaynaklarının erişilebilirliğini kontrol eder.
+    * `search(query)`: ChatGPT Deep Research uyumluluğu için Bedesten destekli kaynaklarda arama yapar.
+    * `fetch(id)`: ChatGPT Deep Research uyumluluğu için tek bir Bedesten belge ID'sinin tam metnini getirir.
+
 </details>
 
 ---
@@ -199,8 +406,8 @@ Bu FastMCP sunucusu **19 optimize edilmiş MCP aracı** sunar (token verimliliğ
 - **Korunan İşlevsellik:** %100 özellik desteği devam ediyor
 
 **GENEL İSTATİSTİKLER:**
-- **Toplam Mahkeme/Kurum:** 13 farklı hukuki kurum (KVKK dahil)
-- **Toplam MCP Tool:** 19 optimize edilmiş arama ve belge getirme aracı  
+- **Toplam Mahkeme/Kurum:** 15 farklı hukuki kurum (GİB Özelgeleri ve Sigorta Tahkim Komisyonu dahil)
+- **Toplam MCP Tool:** 26 aktif araç + 1 opsiyonel semantik arama aracı
 - **Daire/Kurul Filtreleme:** 87 farklı seçenek (52 Yargıtay + 27 Danıştay + 8 Sayıştay)
 - **Tarih Filtreleme:** Birleşik Bedesten API aracında ISO 8601 formatında tam tarih aralığı desteği
 - **Kesin Cümle Arama:** Birleşik Bedesten API aracında çift tırnak ile tam cümle arama (`"\"mülkiyet kararı\""` formatı)

analyze_kik_hash_generation.py

@@ -0,0 +1,239 @@
+#!/usr/bin/env python3
+
+"""
+Analyze KİK v2 hash generation by examining JavaScript code patterns
+and trying to reverse engineer the hash generation logic.
+"""
+
+import asyncio
+import json
+import hashlib
+import hmac
+import base64
+from fastmcp import Client
+from mcp_server_main import app
+
+def analyze_webpack_hash_patterns():
+    """
+    Analyze the webpack JavaScript code you provided to find hash generation patterns
+    """
+    print("🔍 Analyzing webpack hash generation patterns...")
+    
+    # From the JavaScript code, I can see several hash/ID generation patterns:
+    hash_patterns = {
+        # Webpack chunk system hashes (from the JS code)
+        "webpack_chunks": {
+            315: "d9a9486a4f5ba326",
+            531: "cd8fb385c88033ae", 
+            671: "04c48b287646627a",
+            856: "682c9a7b87351f90",
+            1017: "9de022378fc275f6",
+            # ... many more from the __webpack_require__.u function
+        },
+        
+        # Symbol generation from Zone.js
+        "zone_symbols": [
+            "__zone_symbol__",
+            "__Zone_symbol_prefix",
+            "Zone.__symbol__"
+        ],
+        
+        # Angular module federation patterns
+        "module_federation": [
+            "__webpack_modules__",
+            "__webpack_module_cache__",
+            "__webpack_require__"
+        ]
+    }
+    
+    # The target hash format
+    target_hash = "42f9bcd59e0dfbca36dec9accf5686c7a92aa97724cd8fc3550beb84b80409da"
+    print(f"🎯 Target hash: {target_hash}")
+    print(f"   Length: {len(target_hash)} characters")
+    print(f"   Format: {'SHA256' if len(target_hash) == 64 else 'Other'} (64 chars = SHA256)")
+    
+    return hash_patterns
+
+def test_webpack_style_hashing(data_dict):
+    """Test webpack-style hash generation methods"""
+    hashes = {}
+    
+    for key, value in data_dict.items():
+        test_string = str(value)
+        
+        # Try various webpack-style hash methods
+        hashes[f"webpack_md5_{key}"] = hashlib.md5(test_string.encode()).hexdigest()
+        hashes[f"webpack_sha1_{key}"] = hashlib.sha1(test_string.encode()).hexdigest()  
+        hashes[f"webpack_sha256_{key}"] = hashlib.sha256(test_string.encode()).hexdigest()
+        
+        # Try with various prefixes/suffixes (common in webpack)
+        prefixed = f"__webpack__{test_string}"
+        hashes[f"webpack_prefixed_sha256_{key}"] = hashlib.sha256(prefixed.encode()).hexdigest()
+        
+        # Try with module federation style
+        module_style = f"shell:{test_string}"
+        hashes[f"module_fed_sha256_{key}"] = hashlib.sha256(module_style.encode()).hexdigest()
+        
+        # Try JSON stringified
+        json_style = json.dumps({"id": value, "type": "decision"}, separators=(',', ':'))
+        hashes[f"json_sha256_{key}"] = hashlib.sha256(json_style.encode()).hexdigest()
+        
+        # Try with timestamp or sequence
+        with_seq = f"{test_string}_0"
+        hashes[f"seq_sha256_{key}"] = hashlib.sha256(with_seq.encode()).hexdigest()
+        
+    return hashes
+
+def test_angular_routing_hashes(data_dict):
+    """Test Angular routing/state management hash generation"""
+    hashes = {}
+    
+    for key, value in data_dict.items():
+        # Angular often uses route parameters for hash generation
+        route_style = f"/kurul-kararlari/{value}"
+        hashes[f"route_sha256_{key}"] = hashlib.sha256(route_style.encode()).hexdigest()
+        
+        # Component state style
+        state_style = f"KurulKararGoster_{value}"
+        hashes[f"state_sha256_{key}"] = hashlib.sha256(state_style.encode()).hexdigest()
+        
+        # Angular module style
+        module_style = f"kik.kurul.karar.{value}"
+        hashes[f"module_sha256_{key}"] = hashlib.sha256(module_style.encode()).hexdigest()
+        
+    return hashes
+
+def test_base64_encoding_variants(data_dict):
+    """Test various base64 and encoding variants"""
+    hashes = {}
+    
+    for key, value in data_dict.items():
+        test_string = str(value)
+        
+        # Try base64 encoding then hashing
+        b64_encoded = base64.b64encode(test_string.encode()).decode()
+        hashes[f"b64_sha256_{key}"] = hashlib.sha256(b64_encoded.encode()).hexdigest()
+        
+        # Try URL-safe base64
+        b64_url = base64.urlsafe_b64encode(test_string.encode()).decode()
+        hashes[f"b64url_sha256_{key}"] = hashlib.sha256(b64_url.encode()).hexdigest()
+        
+        # Try hex encoding
+        hex_encoded = test_string.encode().hex()
+        hashes[f"hex_sha256_{key}"] = hashlib.sha256(hex_encoded.encode()).hexdigest()
+        
+    return hashes
+
+async def test_hash_generation_comprehensive():
+    print("🔐 Comprehensive KİK document hash generation analysis...")
+    print("=" * 70)
+    
+    # First analyze the webpack patterns
+    webpack_patterns = analyze_webpack_hash_patterns()
+    
+    client = Client(app)
+    
+    async with client:
+        print("✅ MCP client connected")
+        
+        # Get sample decisions 
+        print("\n📊 Getting sample decisions for hash analysis...")
+        search_result = await client.call_tool("search_kik_v2_decisions", {
+            "decision_type": "uyusmazlik",
+            "karar_metni": "2024"
+        })
+        
+        if hasattr(search_result, 'content') and search_result.content:
+            search_data = json.loads(search_result.content[0].text)
+            decisions = search_data.get('decisions', [])
+            
+            if decisions:
+                print(f"✅ Found {len(decisions)} decisions")
+                
+                # Test with first decision
+                sample_decision = decisions[0]
+                print(f"\n📋 Sample decision for hash analysis:")
+                for key, value in sample_decision.items():
+                    print(f"   {key}: {value}")
+                
+                target_hash = "42f9bcd59e0dfbca36dec9accf5686c7a92aa97724cd8fc3550beb84b80409da"
+                print(f"\n🎯 Target hash to match: {target_hash}")
+                
+                all_hashes = {}
+                
+                # Test different hash generation methods
+                print(f"\n🔨 Testing webpack-style hashing...")
+                webpack_hashes = test_webpack_style_hashing(sample_decision)
+                all_hashes.update(webpack_hashes)
+                
+                print(f"🔨 Testing Angular routing hashes...")  
+                angular_hashes = test_angular_routing_hashes(sample_decision)
+                all_hashes.update(angular_hashes)
+                
+                print(f"🔨 Testing base64 encoding variants...")
+                b64_hashes = test_base64_encoding_variants(sample_decision)
+                all_hashes.update(b64_hashes)
+                
+                # Check for matches
+                print(f"\n🎯 Checking for hash matches...")
+                matches_found = []
+                partial_matches = []
+                
+                for hash_name, hash_value in all_hashes.items():
+                    if hash_value == target_hash:
+                        matches_found.append((hash_name, hash_value))
+                        print(f"   🎉 EXACT MATCH FOUND: {hash_name}")
+                    elif hash_value[:8] == target_hash[:8]:  # First 8 chars match
+                        partial_matches.append((hash_name, hash_value))
+                        print(f"   🔍 Partial match (first 8): {hash_name} -> {hash_value[:16]}...")
+                    elif hash_value[-8:] == target_hash[-8:]:  # Last 8 chars match
+                        partial_matches.append((hash_name, hash_value))
+                        print(f"   🔍 Partial match (last 8): {hash_name} -> ...{hash_value[-16:]}")
+                
+                if not matches_found and not partial_matches:
+                    print(f"   ❌ No matches found")
+                    print(f"\n📝 Sample generated hashes (first 10):")
+                    for i, (hash_name, hash_value) in enumerate(list(all_hashes.items())[:10]):
+                        print(f"   {hash_name}: {hash_value}")
+                
+                # Try combinations with other decisions
+                print(f"\n🔄 Testing hash combinations with multiple decisions...")
+                if len(decisions) > 1:
+                    for i, decision in enumerate(decisions[1:3]):  # Test 2 more
+                        print(f"\n   Testing decision {i+2}: {decision.get('kararNo')}")
+                        decision_hashes = test_webpack_style_hashing(decision)
+                        
+                        for hash_name, hash_value in decision_hashes.items():
+                            if hash_value == target_hash:
+                                print(f"   🎉 MATCH FOUND in decision {i+2}: {hash_name}")
+                                matches_found.append((f"decision_{i+2}_{hash_name}", hash_value))
+                
+                # Try composite hashes (combining multiple fields)
+                print(f"\n🔗 Testing composite hash generation...")
+                composite_tests = [
+                    f"{sample_decision.get('gundemMaddesiId')}_{sample_decision.get('kararNo')}",
+                    f"{sample_decision.get('kararNo')}_{sample_decision.get('kararTarihi')}",
+                    f"uyusmazlik_{sample_decision.get('gundemMaddesiId')}_{sample_decision.get('kararTarihi')}",
+                    json.dumps(sample_decision, separators=(',', ':'), sort_keys=True),
+                    f"{sample_decision.get('basvuran')}_{sample_decision.get('gundemMaddesiId')}",
+                ]
+                
+                for i, composite_str in enumerate(composite_tests):
+                    composite_hash = hashlib.sha256(composite_str.encode()).hexdigest()
+                    if composite_hash == target_hash:
+                        print(f"   🎉 COMPOSITE MATCH FOUND: test_{i} -> {composite_str[:50]}...")
+                        matches_found.append((f"composite_{i}", composite_hash))
+                        
+                print(f"\n🎯 Hash analysis completed!")
+                print(f"   Total matches found: {len(matches_found)}")
+                print(f"   Partial matches: {len(partial_matches)}")
+                
+            else:
+                print("❌ No decisions found")
+        else:
+            print("❌ Search failed")
+    
+    print("=" * 70)
+
+if __name__ == "__main__":
+    asyncio.run(test_hash_generation_comprehensive())

anayasa_mcp_module/bireysel_client.py

@@ -1,6 +1,7 @@
 # anayasa_mcp_module/bireysel_client.py
 # This client is for Bireysel Başvuru: https://kararlarbilgibankasi.anayasa.gov.tr
 
+import asyncio
 import httpx
 from bs4 import BeautifulSoup, Tag
 from typing import Dict, Any, List, Optional, Tuple
@@ -302,7 +303,7 @@ class AnayasaBireyselBasvuruApiClient:
                             elif "Karar Tarihi" in key and not karar_tarihi_from_page: karar_tarihi_from_page = value
                             elif "Resmi Gazete Tarih / Sayı" in key: resmi_gazete_info_from_page = value
             
-            full_markdown_content = self._convert_html_to_markdown_bireysel(html_content_from_api)
+            full_markdown_content = await asyncio.to_thread(self._convert_html_to_markdown_bireysel, html_content_from_api)
 
             if not full_markdown_content:
                 return AnayasaBireyselBasvuruDocumentMarkdown(

anayasa_mcp_module/client.py

@@ -1,6 +1,7 @@
 # anayasa_mcp_module/client.py
 # This client is for Norm Denetimi: https://normkararlarbilgibankasi.anayasa.gov.tr
 
+import asyncio
 import httpx
 from bs4 import BeautifulSoup
 from typing import Dict, Any, List, Optional, Tuple
@@ -309,7 +310,7 @@ class AnayasaMahkemesiApiClient:
                     official_gazette_from_page = rg_text_content.replace("Resmî Gazete tarih ve sayısı:", "").replace("Resmi Gazete tarih/sayı:", "").strip()
 
 
-            full_markdown_content = self._convert_html_to_markdown_norm_denetimi(html_content_from_api)
+            full_markdown_content = await asyncio.to_thread(self._convert_html_to_markdown_norm_denetimi, html_content_from_api)
 
             if not full_markdown_content:
                 return AnayasaDocumentMarkdown(

anayasa_mcp_module/unified_client.py

@@ -2,8 +2,8 @@
 # Unified client for both Norm Denetimi and Bireysel Başvuru
 
 import logging
-from typing import Optional
-from urllib.parse import urlparse
+from typing import Optional, Tuple
+from urllib.parse import urlparse, urlunparse
 
 from .models import (
     AnayasaUnifiedSearchRequest,
@@ -18,6 +18,51 @@ from .bireysel_client import AnayasaBireyselBasvuruApiClient
 
 logger = logging.getLogger(__name__)
 
+# Canonical hosts per decision type. Norm Denetimi (/ND/) documents live on the
+# "norm" subdomain; Bireysel Başvuru (/BB/) documents on the plain subdomain.
+# Callers (or upstream search links) sometimes supply the wrong host for a given
+# path, which makes the AYM server return 404. We re-key the host off the path.
+_NORM_HOST = "normkararlarbilgibankasi.anayasa.gov.tr"
+_BIREYSEL_HOST = "kararlarbilgibankasi.anayasa.gov.tr"
+
+
+def normalize_anayasa_document_url(document_url: str) -> Tuple[Optional[str], str]:
+    """Detect the AYM decision type from the URL path and force the correct host.
+
+    Detection is path-based (``/ND/`` vs ``/BB/``) because the path is
+    unambiguous, whereas the supplied host may be wrong. Query params and
+    fragment are preserved (they are harmless for document fetches).
+
+    Returns ``(decision_type, normalized_url)`` where ``decision_type`` is
+    ``"norm_denetimi"``, ``"bireysel_basvuru"``, or ``None`` if it cannot be
+    determined (URL returned unchanged in that case).
+    """
+    parsed = urlparse(document_url)
+    path = parsed.path or ""
+
+    if "/ND/" in path:
+        decision_type, host = "norm_denetimi", _NORM_HOST
+    elif "/BB/" in path:
+        decision_type, host = "bireysel_basvuru", _BIREYSEL_HOST
+    else:
+        # Fall back to host-based detection when the path is uninformative.
+        if "normkararlarbilgibankasi" in parsed.netloc:
+            return "norm_denetimi", document_url
+        if "kararlarbilgibankasi" in parsed.netloc:
+            return "bireysel_basvuru", document_url
+        return None, document_url
+
+    normalized = urlunparse((
+        parsed.scheme or "https",
+        host,
+        parsed.path,
+        parsed.params,
+        parsed.query,
+        parsed.fragment,
+    ))
+    return decision_type, normalized
+
+
 class AnayasaUnifiedClient:
     """Unified client that handles both Norm Denetimi and Bireysel Başvuru searches."""
     
@@ -80,12 +125,18 @@ class AnayasaUnifiedClient:
     async def get_document_unified(self, document_url: str, page_number: int = 1) -> AnayasaUnifiedDocumentMarkdown:
         """Unified document retrieval that auto-detects the appropriate client."""
         
-        # Auto-detect decision type based on URL
-        parsed_url = urlparse(document_url)
+        # Auto-detect decision type from the path and force the correct host.
+        # This repairs malformed URLs (e.g. a /ND/ path on the bireysel host),
+        # which otherwise 404 against the AYM server.
+        decision_type, normalized_url = normalize_anayasa_document_url(document_url)
+        if normalized_url != document_url:
+            logger.info(
+                f"AnayasaUnifiedClient: Normalized document URL "
+                f"'{document_url}' -> '{normalized_url}'"
+            )
 
-        if "normkararlarbilgibankasi" in parsed_url.netloc or "/ND/" in document_url:
-            # Norm Denetimi document
-            result = await self.norm_client.get_decision_document_as_markdown(document_url, page_number)
+        if decision_type == "norm_denetimi":
+            result = await self.norm_client.get_decision_document_as_markdown(normalized_url, page_number)
 
             return AnayasaUnifiedDocumentMarkdown(
                 decision_type="norm_denetimi",
@@ -97,9 +148,8 @@ class AnayasaUnifiedClient:
                 is_paginated=result.is_paginated
             )
 
-        elif "kararlarbilgibankasi" in parsed_url.netloc or "/BB/" in document_url:
-            # Bireysel Başvuru document
-            result = await self.bireysel_client.get_decision_document_as_markdown(document_url, page_number)
+        elif decision_type == "bireysel_basvuru":
+            result = await self.bireysel_client.get_decision_document_as_markdown(normalized_url, page_number)
 
             return AnayasaUnifiedDocumentMarkdown(
                 decision_type="bireysel_basvuru",

app.py

@@ -0,0 +1,35 @@
+"""
+ASGI application for Yargı MCP Server (simple deployment variant).
+
+This is a minimal ASGI application that can be run with:
+    uvicorn app:app --host 0.0.0.0 --port 8000
+
+The MCP server will be available at:
+    http://localhost:8000/mcp/
+
+For the FastAPI-wrapped variant with CORS and extra metadata routes,
+see asgi_app.py instead.
+"""
+
+from starlette.responses import JSONResponse
+from mcp_server_main import create_app
+
+mcp = create_app()
+
+
+@mcp.custom_route("/health", methods=["GET"])
+async def health_check(request):
+    """Health check endpoint for monitoring services (Fly.io, Render, etc.)."""
+    return JSONResponse({
+        "status": "healthy",
+        "service": "Yargı MCP Server",
+        "version": "0.2.1",
+    })
+
+
+# Create ASGI app directly from FastMCP server
+app = mcp.http_app()
+
+# Endpoints:
+# - /mcp/   - MCP server (Streamable HTTP transport, default FastMCP path)
+# - /health - Health check for monitoring

asgi_app.py

@@ -2,109 +2,36 @@
 ASGI application for Yargı MCP Server
 
 This module provides ASGI/HTTP access to the Yargı MCP server,
-allowing it to be deployed as a web service with FastAPI wrapper
-for Stripe webhook integration.
+allowing it to be deployed as a web service with FastAPI wrapper.
 
 Usage:
     uvicorn asgi_app:app --host 0.0.0.0 --port 8000
 """
 
 import os
-import time
+import json
 import logging
-from datetime import datetime, timedelta
-from fastapi import FastAPI, Request, HTTPException, Query
-from fastapi.responses import JSONResponse, HTMLResponse
-from fastapi.exception_handlers import http_exception_handler
+from fastapi import FastAPI, Request
+from fastapi.responses import JSONResponse
 from starlette.middleware import Middleware
 from starlette.middleware.cors import CORSMiddleware
-from starlette.responses import Response
-from starlette.requests import Request as StarletteRequest
 
-# Import the MCP app creator function
 from mcp_server_main import create_app
 
-# Import Stripe webhook router
-from stripe_webhook import router as stripe_router
-
-# Import simplified MCP Auth HTTP adapter
-from mcp_auth_http_simple import router as mcp_auth_router
-
-# OAuth configuration from environment variables
-CLERK_ISSUER = os.getenv("CLERK_ISSUER", "https://accounts.yargimcp.com")
-BASE_URL = os.getenv("BASE_URL", "https://yargimcp.com")
-
 # Setup logging
 logger = logging.getLogger(__name__)
 
-# Configure CORS and Auth middleware
+# Configure CORS
 cors_origins = os.getenv("ALLOWED_ORIGINS", "*").split(",")
 
-# Import FastMCP Bearer Auth Provider
-from fastmcp.server.auth import BearerAuthProvider
-from fastmcp.server.auth.providers.bearer import RSAKeyPair
+# Create MCP app
+mcp_server = create_app()
 
-# Clerk JWT configuration for Bearer token validation
-CLERK_SECRET_KEY = os.getenv("CLERK_SECRET_KEY")
-CLERK_ISSUER = os.getenv("CLERK_ISSUER", "https://accounts.yargimcp.com")
-CLERK_PUBLISHABLE_KEY = os.getenv("CLERK_PUBLISHABLE_KEY")
-
-# Configure Bearer token authentication
-bearer_auth = None
-if CLERK_SECRET_KEY and CLERK_ISSUER:
-    # Production: Use Clerk JWKS endpoint for token validation
-    bearer_auth = BearerAuthProvider(
-        jwks_uri=f"{CLERK_ISSUER}/.well-known/jwks.json",
-        issuer=CLERK_ISSUER,
-        algorithm="RS256",
-        audience=None,  # Disable audience validation - Clerk uses different audience format
-        required_scopes=[]  # Disable scope validation - Clerk JWT has ['read', 'search']
-    )
-    logger.info(f"Bearer auth configured with Clerk JWKS: {CLERK_ISSUER}/.well-known/jwks.json")
-else:
-    # Development: Generate RSA key pair for testing
-    logger.warning("No Clerk credentials found - using development RSA key pair")
-    dev_key_pair = RSAKeyPair.generate()
-    bearer_auth = BearerAuthProvider(
-        public_key=dev_key_pair.public_key,
-        issuer="https://dev.yargimcp.com",
-        audience="dev-mcp-server",
-        required_scopes=["yargi.read"]
-    )
-    
-    # Generate a test token for development
-    dev_token = dev_key_pair.create_token(
-        subject="dev-user",
-        issuer="https://dev.yargimcp.com",
-        audience="dev-mcp-server",
-        scopes=["yargi.read", "yargi.search"],
-        expires_in_seconds=3600 * 24  # 24 hours for development
-    )
-    logger.info(f"Development Bearer token: {dev_token}")
-
-custom_middleware = [
-    Middleware(
-        CORSMiddleware,
-        allow_origins=cors_origins,
-        allow_credentials=True,
-        allow_methods=["GET", "POST", "OPTIONS", "DELETE"],
-        allow_headers=["Content-Type", "Authorization", "X-Request-ID", "X-Session-ID"],
-    ),
-]
-
-# Create MCP app with Bearer authentication
-mcp_server = create_app(auth=bearer_auth)
-
-# Add Starlette middleware to FastAPI (not MCP)
-# MCP already has Bearer auth, no need for additional middleware on MCP level
-
-# Create MCP Starlette sub-application with root path - mount will add /mcp prefix
+# Create MCP Starlette sub-application
 mcp_app = mcp_server.http_app(path="/")
 
-# Configure JSON encoder for proper Turkish character support
-import json
-from fastapi.responses import JSONResponse
 
+# Configure JSON encoder for proper Turkish character support
 class UTF8JSONResponse(JSONResponse):
     def __init__(self, content=None, status_code=200, headers=None, **kwargs):
         if headers is None:
@@ -121,82 +48,55 @@ class UTF8JSONResponse(JSONResponse):
             separators=(",", ":"),
         ).encode("utf-8")
 
+custom_middleware = [
+    Middleware(
+        CORSMiddleware,
+        allow_origins=cors_origins,
+        allow_credentials=True,
+        allow_methods=["GET", "POST", "OPTIONS"],
+        allow_headers=["Content-Type", "X-Request-ID", "X-Session-ID"],
+    ),
+]
+
 # Create FastAPI wrapper application
 app = FastAPI(
     title="Yargı MCP Server",
-    description="MCP server for Turkish legal databases with OAuth authentication",
+    description="MCP server for Turkish legal databases",
     version="0.1.0",
     middleware=custom_middleware,
-    default_response_class=UTF8JSONResponse  # Use UTF-8 JSON encoder
+    default_response_class=UTF8JSONResponse,
+    redirect_slashes=False,
 )
 
-# Add Stripe webhook router to FastAPI
-app.include_router(stripe_router, prefix="/api")
 
-# Add MCP Auth HTTP adapter to FastAPI (handles OAuth endpoints)
-app.include_router(mcp_auth_router)
-
-# Custom 401 exception handler for MCP spec compliance
-@app.exception_handler(401)
-async def custom_401_handler(request: Request, exc: HTTPException):
-    """Custom 401 handler that adds WWW-Authenticate header as required by MCP spec"""
-    response = await http_exception_handler(request, exc)
-    
-    # Add WWW-Authenticate header pointing to protected resource metadata
-    # as required by RFC 9728 Section 5.1 and MCP Authorization spec
-    response.headers["WWW-Authenticate"] = (
-        'Bearer '
-        'error="invalid_token", '
-        'error_description="The access token is missing or invalid", '
-        f'resource="{BASE_URL}/.well-known/oauth-protected-resource"'
-    )
-    
-    return response
-
-# FastAPI health check endpoint - BEFORE mounting MCP app
 @app.get("/health")
 async def health_check():
     """Health check endpoint for monitoring"""
-    return JSONResponse({
+    return {
         "status": "healthy",
         "service": "Yargı MCP Server",
         "version": "0.1.0",
         "tools_count": len(mcp_server._tool_manager._tools),
-        "auth_enabled": os.getenv("ENABLE_AUTH", "false").lower() == "true"
-    })
+    }
+
 
-# Add explicit redirect for /mcp to /mcp/ with method preservation
 @app.api_route("/mcp", methods=["GET", "POST", "HEAD", "OPTIONS"])
 async def redirect_to_slash(request: Request):
     """Redirect /mcp to /mcp/ preserving HTTP method with 308"""
     from fastapi.responses import RedirectResponse
     return RedirectResponse(url="/mcp/", status_code=308)
 
-# Mount MCP app at /mcp/ with trailing slash
-app.mount("/mcp/", mcp_app)
 
-# Set the lifespan context after mounting
-app.router.lifespan_context = mcp_app.lifespan
-
-
-# SSE transport deprecated - removed
-
-# FastAPI root endpoint
 @app.get("/")
 async def root():
     """Root endpoint with service information"""
-    return JSONResponse({
+    return {
         "service": "Yargı MCP Server",
-        "description": "MCP server for Turkish legal databases with OAuth authentication",
+        "description": "MCP server for Turkish legal databases",
         "endpoints": {
             "mcp": "/mcp",
             "health": "/health",
             "status": "/status",
-            "stripe_webhook": "/api/stripe/webhook",
-            "oauth_login": "/auth/login",
-            "oauth_callback": "/auth/callback",
-            "oauth_google": "/auth/google/login",
-            "user_info": "/auth/user"
         },
         "transports": {
             "http": "/mcp"
@@ -210,166 +110,14 @@ async def root():
             "Kamu İhale Kurulu (Public Procurement Authority)",
             "Rekabet Kurumu (Competition Authority)",
             "Sayıştay (Court of Accounts)",
-            "Bedesten API (Multiple courts)"
+            "KVKK (Personal Data Protection Authority)",
+            "BDDK (Banking Regulation and Supervision Agency)",
+            "Bedesten API (Multiple courts)",
+            "Sigorta Tahkim Komisyonu (Insurance Arbitration Commission)",
         ],
-        "authentication": {
-            "enabled": os.getenv("ENABLE_AUTH", "false").lower() == "true",
-            "type": "OAuth 2.0 via Clerk",
-            "issuer": os.getenv("CLERK_ISSUER", "https://clerk.accounts.dev"),
-            "providers": ["google"],
-            "flow": "authorization_code"
     }
-    })
 
-# OAuth 2.0 Authorization Server Metadata proxy (for MCP clients that can't reach Clerk directly)
-# MCP Auth Toolkit expects this to be under /mcp/.well-known/oauth-authorization-server
-@app.get("/mcp/.well-known/oauth-authorization-server")
-async def oauth_authorization_server():
-    """OAuth 2.0 Authorization Server Metadata proxy to Clerk - MCP Auth Toolkit standard location"""
-    return JSONResponse({
-        "issuer": BASE_URL,
-        "authorization_endpoint": "https://yargimcp.com/mcp-callback",
-        "token_endpoint": f"{BASE_URL}/token", 
-        "jwks_uri": f"{CLERK_ISSUER}/.well-known/jwks.json",
-        "response_types_supported": ["code"],
-        "grant_types_supported": ["authorization_code", "refresh_token"],
-        "token_endpoint_auth_methods_supported": ["client_secret_basic", "none"],
-        "scopes_supported": ["read", "search", "openid", "profile", "email"],
-        "subject_types_supported": ["public"],
-        "id_token_signing_alg_values_supported": ["RS256"],
-        "claims_supported": ["sub", "iss", "aud", "exp", "iat", "email", "name"],
-        "code_challenge_methods_supported": ["S256"],
-        "service_documentation": f"{BASE_URL}/mcp",
-        "registration_endpoint": f"{BASE_URL}/register",
-        "resource_documentation": f"{BASE_URL}/mcp"
-    })
 
-# Claude AI MCP specific endpoint format
-@app.get("/.well-known/oauth-authorization-server/mcp")
-async def oauth_authorization_server_mcp_suffix():
-    """OAuth 2.0 Authorization Server Metadata - Claude AI MCP specific format"""
-    return JSONResponse({
-        "issuer": BASE_URL,
-        "authorization_endpoint": "https://yargimcp.com/mcp-callback",
-        "token_endpoint": f"{BASE_URL}/token", 
-        "jwks_uri": f"{CLERK_ISSUER}/.well-known/jwks.json",
-        "response_types_supported": ["code"],
-        "grant_types_supported": ["authorization_code", "refresh_token"],
-        "token_endpoint_auth_methods_supported": ["client_secret_basic", "none"],
-        "scopes_supported": ["read", "search", "openid", "profile", "email"],
-        "subject_types_supported": ["public"],
-        "id_token_signing_alg_values_supported": ["RS256"],
-        "claims_supported": ["sub", "iss", "aud", "exp", "iat", "email", "name"],
-        "code_challenge_methods_supported": ["S256"],
-        "service_documentation": f"{BASE_URL}/mcp",
-        "registration_endpoint": f"{BASE_URL}/register",
-        "resource_documentation": f"{BASE_URL}/mcp"
-    })
-
-@app.get("/.well-known/oauth-protected-resource/mcp")
-async def oauth_protected_resource_mcp_suffix():
-    """OAuth 2.0 Protected Resource Metadata - Claude AI MCP specific format"""
-    return JSONResponse({
-        "resource": BASE_URL,
-        "authorization_servers": [
-            BASE_URL
-        ],
-        "scopes_supported": ["read", "search"],
-        "bearer_methods_supported": ["header"],
-        "resource_documentation": f"{BASE_URL}/mcp",
-        "resource_policy_uri": f"{BASE_URL}/privacy"
-    })
-
-# Keep root level for compatibility with some MCP clients
-@app.get("/.well-known/oauth-authorization-server")
-async def oauth_authorization_server_root():
-    """OAuth 2.0 Authorization Server Metadata proxy to Clerk - root level for compatibility"""
-    return JSONResponse({
-        "issuer": BASE_URL,
-        "authorization_endpoint": "https://yargimcp.com/mcp-callback",
-        "token_endpoint": f"{BASE_URL}/token", 
-        "jwks_uri": f"{CLERK_ISSUER}/.well-known/jwks.json",
-        "response_types_supported": ["code"],
-        "grant_types_supported": ["authorization_code", "refresh_token"],
-        "token_endpoint_auth_methods_supported": ["client_secret_basic", "none"],
-        "scopes_supported": ["read", "search", "openid", "profile", "email"],
-        "subject_types_supported": ["public"],
-        "id_token_signing_alg_values_supported": ["RS256"],
-        "claims_supported": ["sub", "iss", "aud", "exp", "iat", "email", "name"],
-        "code_challenge_methods_supported": ["S256"],
-        "service_documentation": f"{BASE_URL}/mcp",
-        "registration_endpoint": f"{BASE_URL}/register",
-        "resource_documentation": f"{BASE_URL}/mcp"
-    })
-
-# Note: GET /mcp is handled by the mounted MCP app itself
-# This prevents 405 Method Not Allowed errors on POST requests
-
-# OAuth 2.0 Protected Resource Metadata (RFC 9728) - MCP Spec Required
-@app.get("/.well-known/oauth-protected-resource")
-async def oauth_protected_resource():
-    """OAuth 2.0 Protected Resource Metadata as required by MCP spec"""
-    return JSONResponse({
-        "resource": BASE_URL,
-        "authorization_servers": [
-            BASE_URL
-        ],
-        "scopes_supported": ["read", "search"],
-        "bearer_methods_supported": ["header"],
-        "resource_documentation": f"{BASE_URL}/mcp",
-        "resource_policy_uri": f"{BASE_URL}/privacy"
-    })
-
-# Standard well-known discovery endpoint
-@app.get("/.well-known/mcp")
-async def well_known_mcp():
-    """Standard MCP discovery endpoint"""
-    return JSONResponse({
-        "mcp_server": {
-            "name": "Yargı MCP Server",
-            "version": "0.1.0",
-            "endpoint": f"{BASE_URL}/mcp",
-            "authentication": {
-                "type": "oauth2",
-                "authorization_url": f"{BASE_URL}/auth/login",
-                "scopes": ["read", "search"]
-            },
-            "capabilities": ["tools", "resources"],
-            "tools_count": len(mcp_server._tool_manager._tools)
-        }
-    })
-
-# MCP Discovery endpoint for ChatGPT integration
-@app.get("/mcp/discovery")
-async def mcp_discovery():
-    """MCP Discovery endpoint for ChatGPT and other MCP clients"""
-    return JSONResponse({
-        "name": "Yargı MCP Server",
-        "description": "MCP server for Turkish legal databases",
-        "version": "0.1.0",
-        "protocol": "mcp",
-        "transport": "http",
-        "endpoint": "/mcp",
-        "authentication": {
-            "type": "oauth2",
-            "authorization_url": "/auth/login",
-            "token_url": "/auth/callback",
-            "scopes": ["read", "search"],
-            "provider": "clerk"
-        },
-        "capabilities": {
-            "tools": True,
-            "resources": True,
-            "prompts": False
-        },
-        "tools_count": len(mcp_server._tool_manager._tools),
-        "contact": {
-            "url": BASE_URL,
-            "email": "support@yargi-mcp.dev"
-        }
-    })
-
-# FastAPI status endpoint
 @app.get("/status")
 async def status():
     """Status endpoint with detailed information"""
@@ -380,187 +128,19 @@ async def status():
             "description": tool.description[:100] + "..." if len(tool.description) > 100 else tool.description
         })
 
-    return JSONResponse({
+    return {
         "status": "operational",
         "tools": tools,
         "total_tools": len(tools),
         "transport": "streamable_http",
-        "architecture": "FastAPI wrapper + MCP Starlette sub-app",
-        "auth_status": "enabled" if os.getenv("ENABLE_AUTH", "false").lower() == "true" else "disabled"
-    })
+    }
 
-# Note: JWT token validation is now handled entirely by Clerk
-# All authentication flows use Clerk JWT tokens directly
 
-async def validate_clerk_session(request: Request, clerk_token: str = None) -> str:
-    """Validate Clerk session from cookies or JWT token and return user_id"""
-    logger.info(f"Validating Clerk session - token provided: {bool(clerk_token)}")
+# Mount MCP app at /mcp/
+app.mount("/mcp/", mcp_app)
 
-    try:
-        # Try to import Clerk SDK
-        from clerk_backend_api import Clerk
-        clerk = Clerk(bearer_auth=os.getenv("CLERK_SECRET_KEY"))
-        
-        # Try JWT token first (from URL parameter)
-        if clerk_token:
-            logger.info("Validating Clerk JWT token from URL parameter")
-            try:
-                # Extract session_id from JWT token and verify with Clerk
-                import jwt
-                decoded_token = jwt.decode(clerk_token, options={"verify_signature": False})
-                session_id = decoded_token.get("sid")  # Use standard JWT 'sid' claim
-                
-                if session_id:
-                    # Verify with Clerk using session_id
-                    session = clerk.sessions.verify(session_id=session_id, token=clerk_token)
-                    user_id = session.user_id if session else None
-                    
-                    if user_id:
-                        logger.info(f"JWT token validation successful - user_id: {user_id}")
-                        return user_id
-                    else:
-                        logger.error("JWT token validation failed - no user_id in session")
-                else:
-                    logger.error("No session_id found in JWT token")
-            except Exception as e:
-                logger.error(f"JWT token validation failed: {str(e)}")
-                # Fall through to cookie validation
-        
-        # Fallback to cookie validation
-        logger.info("Attempting cookie-based session validation")
-        clerk_session = request.cookies.get("__session")
-        if not clerk_session:
-            logger.error("No Clerk session cookie found")
-            raise HTTPException(status_code=401, detail="No Clerk session found")
-        
-        # Validate session with Clerk
-        session = clerk.sessions.verify_session(clerk_session)
-        logger.info(f"Cookie session validation successful - user_id: {session.user_id}")
-        return session.user_id
-        
-    except ImportError:
-        # Fallback for development without Clerk SDK
-        logger.warning("Clerk SDK not available - using development fallback")
-        return "dev_user_123"
-    except Exception as e:
-        logger.error(f"Session validation failed: {str(e)}")
-        raise HTTPException(status_code=401, detail=f"Session validation failed: {str(e)}")
-
-# MCP OAuth Callback Endpoint
-@app.get("/auth/mcp-callback")
-async def mcp_oauth_callback(request: Request, clerk_token: str = Query(None)):
-    """Handle OAuth callback for MCP token generation"""
-    logger.info(f"MCP OAuth callback - clerk_token provided: {bool(clerk_token)}")
-    
-    try:
-        # Validate Clerk session with JWT token support
-        user_id = await validate_clerk_session(request, clerk_token)
-        logger.info(f"User authenticated successfully - user_id: {user_id}")
-        
-        # Use the Clerk JWT token directly (no need to generate custom token)
-        logger.info("User authenticated successfully via Clerk")
-        
-        # Return success response
-        return HTMLResponse(f"""
-        <html>
-            <head>
-                <title>MCP Connection Successful</title>
-                <style>
-                    body {{ font-family: Arial, sans-serif; text-align: center; padding: 50px; }}
-                    .success {{ color: #28a745; }}
-                    .token {{ background: #f8f9fa; padding: 15px; border-radius: 5px; margin: 20px 0; word-break: break-all; }}
-                </style>
-            </head>
-            <body>
-                <h1 class="success">✅ MCP Connection Successful!</h1>
-                <p>Your Yargı MCP integration is now active.</p>
-                <div class="token">
-                    <strong>Authentication:</strong><br>
-                    <code>Use your Clerk JWT token directly with Bearer authentication</code>
-                </div>
-                <p>You can now close this window and return to your MCP client.</p>
-                <script>
-                    // Try to close the popup if opened as such
-                    if (window.opener) {{
-                        window.opener.postMessage({{
-                            type: 'MCP_AUTH_SUCCESS',
-                            token: 'use_clerk_jwt_token'
-                        }}, '*');
-                        setTimeout(() => window.close(), 3000);
-                    }}
-                </script>
-            </body>
-        </html>
-        """)
-        
-    except HTTPException as e:
-        logger.error(f"MCP OAuth callback failed: {e.detail}")
-        return HTMLResponse(f"""
-        <html>
-            <head>
-                <title>MCP Connection Failed</title>
-                <style>
-                    body {{ font-family: Arial, sans-serif; text-align: center; padding: 50px; }}
-                    .error {{ color: #dc3545; }}
-                    .debug {{ background: #f8f9fa; padding: 10px; margin: 20px 0; border-radius: 5px; font-family: monospace; }}
-                </style>
-            </head>
-            <body>
-                <h1 class="error">❌ MCP Connection Failed</h1>
-                <p>{e.detail}</p>
-                <div class="debug">
-                    <strong>Debug Info:</strong><br>
-                    Clerk Token: {'✅ Provided' if clerk_token else '❌ Missing'}<br>
-                    Error: {e.detail}<br>
-                    Status: {e.status_code}
-                </div>
-                <p>Please try again or contact support.</p>
-                <a href="https://yargimcp.com/sign-in">Return to Sign In</a>
-            </body>
-        </html>
-        """, status_code=e.status_code)
-    except Exception as e:
-        logger.error(f"Unexpected error in MCP OAuth callback: {str(e)}")
-        return HTMLResponse(f"""
-        <html>
-            <head>
-                <title>MCP Connection Error</title>
-                <style>
-                    body {{ font-family: Arial, sans-serif; text-align: center; padding: 50px; }}
-                    .error {{ color: #dc3545; }}
-                </style>
-            </head>
-            <body>
-                <h1 class="error">❌ Unexpected Error</h1>
-                <p>An unexpected error occurred during authentication.</p>
-                <p>Error: {str(e)}</p>
-                <a href="https://yargimcp.com/sign-in">Return to Sign In</a>
-            </body>
-        </html>
-        """, status_code=500)
-
-# OAuth2 Token Endpoint - Now uses Clerk JWT tokens directly
-@app.post("/auth/mcp-token")
-async def mcp_token_endpoint(request: Request):
-    """OAuth2 token endpoint for MCP clients - returns Clerk JWT token info"""
-    try:
-        # Validate Clerk session
-        user_id = await validate_clerk_session(request)
-        
-        return JSONResponse({
-            "message": "Use your Clerk JWT token directly with Bearer authentication",
-            "token_type": "Bearer",
-            "scope": "yargi.read",
-            "user_id": user_id,
-            "instructions": "Include 'Authorization: Bearer YOUR_CLERK_JWT_TOKEN' in your requests"
-        })
-    except HTTPException as e:
-        return JSONResponse(
-            status_code=e.status_code,
-            content={"error": "invalid_request", "error_description": e.detail}
-        )
-
-# Note: Only HTTP transport supported - SSE transport deprecated
+# Set the lifespan context after mounting
+app.router.lifespan_context = mcp_app.lifespan
 
 # Export for uvicorn
 __all__ = ["app"]

bddk_mcp_module/client.py

@@ -1,5 +1,6 @@
 # bddk_mcp_module/client.py
 
+import asyncio
 import httpx
 from typing import List, Optional, Dict, Any
 import logging
@@ -210,14 +211,19 @@ class BddkApiClient:
             
             # Convert to Markdown based on content type
             if "pdf" in content_type:
-                # Handle PDF documents
+                # Handle PDF documents. markitdown is sync; offload to thread
+                # so PDF parsing doesn't block the event-loop / other requests.
                 pdf_stream = io.BytesIO(response.content)
-                result = self.markitdown.convert_stream(pdf_stream, file_extension=".pdf")
+                result = await asyncio.to_thread(
+                    self.markitdown.convert_stream, pdf_stream, file_extension=".pdf"
+                )
                 markdown_content = result.text_content
             else:
-                # Handle HTML documents
+                # Handle HTML documents (sync conversion offloaded to thread)
                 html_stream = io.BytesIO(response.content)
-                result = self.markitdown.convert_stream(html_stream, file_extension=".html")
+                result = await asyncio.to_thread(
+                    self.markitdown.convert_stream, html_stream, file_extension=".html"
+                )
                 markdown_content = result.text_content
             
             # Clean up the markdown content

bedesten_mcp_module/client.py

@@ -1,11 +1,15 @@
 # bedesten_mcp_module/client.py
 
-import httpx
+import asyncio
 import base64
-from typing import Optional
-import logging
-from markitdown import MarkItDown
 import io
+import logging
+import os
+import time
+from typing import Optional
+
+import httpx
+from markitdown import MarkItDown
 
 from .models import (
     BedestenSearchRequest, BedestenSearchResponse,
@@ -16,6 +20,72 @@ from .enums import get_full_birim_adi
 
 logger = logging.getLogger(__name__)
 
+
+class BedestenRateLimited(Exception):
+    """Raised when the local rate-limit bucket would block longer than allowed.
+
+    Carries the suggested retry-after (seconds) so callers can surface a
+    structured 429-style response to the MCP client instead of silently
+    blocking the event-loop slot for the full bucket-pause window.
+    """
+
+    def __init__(self, retry_after: float) -> None:
+        self.retry_after = retry_after
+        super().__init__(f"local bucket would block {retry_after:.1f}s")
+
+
+class _TokenBucket:
+    """Asyncio token bucket with explicit back-pressure.
+
+    Measured Bedesten limit (per source IP, 2026-05-08): 10 requests per
+    rolling 30s window with full refill — equivalent to capacity=10,
+    refill_rate=1 token / 3s. Even with margin, 429s still leak through
+    when other clients share the egress IP, so we also expose
+    ``penalize_until`` so callers can freeze the bucket when the server
+    actually returns 429 (Retry-After).
+    """
+
+    def __init__(self, capacity: int, refill_per_s: float) -> None:
+        self.capacity = float(capacity)
+        self.refill_per_s = float(refill_per_s)
+        self._tokens = float(capacity)
+        self._last = time.monotonic()
+        self._not_before = 0.0
+        self._lock = asyncio.Lock()
+
+    async def acquire(self, max_wait: Optional[float] = None) -> None:
+        """Acquire one token. If ``max_wait`` is set and the next wait would
+        exceed it, raise :class:`BedestenRateLimited` immediately instead of
+        sleeping — keeps a single rate-limited request from holding the
+        worker-slot for the full bucket-pause window (up to ~30s on 429)."""
+        deadline = (time.monotonic() + max_wait) if max_wait is not None else None
+        while True:
+            async with self._lock:
+                now = time.monotonic()
+                if now < self._not_before:
+                    wait_s = self._not_before - now
+                else:
+                    self._tokens = min(
+                        self.capacity,
+                        self._tokens + (now - self._last) * self.refill_per_s,
+                    )
+                    self._last = now
+                    if self._tokens >= 1.0:
+                        self._tokens -= 1.0
+                        return
+                    wait_s = (1.0 - self._tokens) / self.refill_per_s
+            if deadline is not None:
+                remaining = deadline - time.monotonic()
+                if wait_s > remaining:
+                    raise BedestenRateLimited(retry_after=wait_s)
+            await asyncio.sleep(wait_s)
+
+    def penalize_until(self, monotonic_deadline: float) -> None:
+        """Pause the bucket until ``monotonic_deadline`` (drains tokens)."""
+        self._not_before = max(self._not_before, monotonic_deadline)
+        self._tokens = 0.0
+        self._last = time.monotonic()
+
 class BedestenApiClient:
     """
     API Client for Bedesten (bedesten.adalet.gov.tr) - Alternative legal decision search system.
@@ -25,6 +95,17 @@ class BedestenApiClient:
     SEARCH_ENDPOINT = "/emsal-karar/searchDocuments"
     DOCUMENT_ENDPOINT = "/emsal-karar/getDocumentContent"
     
+    # Measured limit (per source IP): 10 requests per 30s window with full
+    # refill (≈ 1 token / 3s steady). We default to 1-token capacity and
+    # 3.5s spacing (no burst, ~14% safety margin). Override via env:
+    #   BEDESTEN_RATE_CAPACITY (default 1)
+    #   BEDESTEN_RATE_REFILL_S (default 3.5; seconds per token)
+    #   BEDESTEN_RATE_MAX_WAIT_S (default 8.0; max seconds to wait in the
+    #     local bucket before returning a structured 429 to the caller)
+    _DEFAULT_CAPACITY = int(os.getenv("BEDESTEN_RATE_CAPACITY", "1"))
+    _DEFAULT_REFILL_S = float(os.getenv("BEDESTEN_RATE_REFILL_S", "3.5"))
+    _DEFAULT_MAX_WAIT_S = float(os.getenv("BEDESTEN_RATE_MAX_WAIT_S", "8.0"))
+
     def __init__(self, request_timeout: float = 60.0):
         self.http_client = httpx.AsyncClient(
             base_url=self.BASE_URL,
@@ -42,6 +123,24 @@ class BedestenApiClient:
             },
             timeout=request_timeout
         )
+        self._bucket = _TokenBucket(
+            capacity=self._DEFAULT_CAPACITY,
+            refill_per_s=1.0 / self._DEFAULT_REFILL_S,
+        )
+
+    def _handle_429(self, response: httpx.Response, op: str) -> None:
+        """Apply back-pressure to the shared bucket based on Retry-After."""
+        retry_after_raw = response.headers.get("Retry-After", "")
+        try:
+            retry_after = float(retry_after_raw)
+        except (TypeError, ValueError):
+            retry_after = 30.0
+        # Cap penalty so a hostile/buggy server can't freeze us indefinitely.
+        retry_after = max(1.0, min(retry_after, 60.0))
+        self._bucket.penalize_until(time.monotonic() + retry_after + 0.5)
+        logger.warning(
+            f"BedestenApiClient: 429 on {op}; bucket paused {retry_after + 0.5:.1f}s"
+        )
     
     async def search_documents(self, search_request: BedestenSearchRequest) -> BedestenSearchResponse:
         """
@@ -63,10 +162,13 @@ class BedestenApiClient:
             if not request_dict["data"]["birimAdi"]:  # Remove if empty string
                 del request_dict["data"]["birimAdi"]
             
+            await self._bucket.acquire(max_wait=self._DEFAULT_MAX_WAIT_S)
             response = await self.http_client.post(
                 self.SEARCH_ENDPOINT,
                 json=request_dict
             )
+            if response.status_code == 429:
+                self._handle_429(response, "search")
             response.raise_for_status()
             response_json = response.json()
 
@@ -94,26 +196,51 @@ class BedestenApiClient:
             )
             
             # Get document
+            await self._bucket.acquire(max_wait=self._DEFAULT_MAX_WAIT_S)
             response = await self.http_client.post(
                 self.DOCUMENT_ENDPOINT,
                 json=doc_request.model_dump()
             )
+            if response.status_code == 429:
+                self._handle_429(response, f"document {document_id}")
             response.raise_for_status()
             response_json = response.json()
             doc_response = BedestenDocumentResponse(**response_json)
             
-            # Decode base64 content
+            # Add null safety checks for document data
+            if not hasattr(doc_response, 'data') or doc_response.data is None:
+                raise ValueError("Document response does not contain data")
+            
+            if not hasattr(doc_response.data, 'content') or doc_response.data.content is None:
+                raise ValueError("Document data does not contain content")
+                
+            if not hasattr(doc_response.data, 'mimeType') or doc_response.data.mimeType is None:
+                raise ValueError("Document data does not contain mimeType")
+            
+            # Decode base64 content with error handling
+            try:
                 content_bytes = base64.b64decode(doc_response.data.content)
+            except Exception as e:
+                raise ValueError(f"Failed to decode base64 content: {str(e)}")
+            
             mime_type = doc_response.data.mimeType
             
             logger.info(f"BedestenApiClient: Document mime type: {mime_type}")
             
-            # Convert to markdown based on mime type
+            # Convert to markdown based on mime type. markitdown is sync and
+            # PDF parsing in particular can block the event-loop for seconds,
+            # which on a single-worker uvicorn deployment stalls every other
+            # in-flight MCP request and new TLS handshakes. Offload to a
+            # thread so the event-loop stays responsive.
             if mime_type == "text/html":
                 html_content = content_bytes.decode('utf-8')
-                markdown_content = self._convert_html_to_markdown(html_content)
+                markdown_content = await asyncio.to_thread(
+                    self._convert_html_to_markdown, html_content
+                )
             elif mime_type == "application/pdf":
-                markdown_content = self._convert_pdf_to_markdown(content_bytes)
+                markdown_content = await asyncio.to_thread(
+                    self._convert_pdf_to_markdown, content_bytes
+                )
             else:
                 logger.warning(f"Unsupported mime type: {mime_type}")
                 markdown_content = f"Unsupported content type: {mime_type}. Unable to convert to markdown."
@@ -121,7 +248,7 @@ class BedestenApiClient:
             return BedestenDocumentMarkdown(
                 documentId=document_id,
                 markdown_content=markdown_content,
-                source_url=f"{self.BASE_URL}/document/{document_id}",
+                source_url=f"https://mevzuat.adalet.gov.tr/ictihat/{document_id}",
                 mime_type=mime_type
             )
             

danistay_mcp_module/client.py

@@ -1,5 +1,6 @@
 # danistay_mcp_module/client.py
 
+import asyncio
 import httpx
 from bs4 import BeautifulSoup
 from typing import Dict, Any, List, Optional
@@ -170,7 +171,7 @@ class DanistayApiClient:
                     source_url=source_url
                 )
 
-            markdown_content = self._convert_html_to_markdown_danistay(html_content_from_api)
+            markdown_content = await asyncio.to_thread(self._convert_html_to_markdown_danistay, html_content_from_api)
 
             return DanistayDocumentMarkdown(
                 id=id,

docker-compose.yml

@@ -1,66 +0,0 @@
-version: '3.8'
-
-services:
-  yargi-mcp:
-    build: .
-    image: yargi-mcp:latest
-    container_name: yargi-mcp-server
-    ports:
-      - "${PORT:-8000}:8000"
-    environment:
-      - HOST=0.0.0.0
-      - PORT=8000
-      - LOG_LEVEL=${LOG_LEVEL:-info}
-      - ALLOWED_ORIGINS=${ALLOWED_ORIGINS:-*}
-      - API_TOKEN=${API_TOKEN:-}
-      - PYTHONUNBUFFERED=1
-    volumes:
-      # Mount logs directory
-      - ./logs:/app/logs
-      # Mount .env file if it exists
-      - ./.env:/app/.env:ro
-    restart: unless-stopped
-    healthcheck:
-      test: ["CMD", "python", "-c", "import httpx; httpx.get('http://localhost:8000/health').raise_for_status()"]
-      interval: 30s
-      timeout: 10s
-      retries: 3
-      start_period: 10s
-    networks:
-      - yargi-network
-
-  # Optional: Nginx reverse proxy
-  nginx:
-    image: nginx:alpine
-    container_name: yargi-nginx
-    ports:
-      - "80:80"
-      - "443:443"
-    volumes:
-      - ./nginx.conf:/etc/nginx/nginx.conf:ro
-      - ./ssl:/etc/nginx/ssl:ro
-    depends_on:
-      - yargi-mcp
-    networks:
-      - yargi-network
-    profiles:
-      - production
-
-  # Optional: Redis for caching (future enhancement)
-  redis:
-    image: redis:alpine
-    container_name: yargi-redis
-    command: redis-server --appendonly yes
-    volumes:
-      - redis-data:/data
-    networks:
-      - yargi-network
-    profiles:
-      - with-cache
-
-networks:
-  yargi-network:
-    driver: bridge
-
-volumes:
-  redis-data:

docs/DEPLOYMENT.md

@@ -1,428 +0,0 @@
-# Yargı MCP Server Dağıtım Rehberi
-
-Bu rehber, Yargı MCP Server'ın ASGI web servisi olarak çeşitli dağıtım seçeneklerini kapsar.
-
-## İçindekiler
-
-- [Hızlı Başlangıç](#hızlı-başlangıç)
-- [Yerel Geliştirme](#yerel-geliştirme)
-- [Production Dağıtımı](#production-dağıtımı)
-- [Cloud Dağıtımı](#cloud-dağıtımı)
-- [Docker Dağıtımı](#docker-dağıtımı)
-- [Güvenlik Hususları](#güvenlik-hususları)
-- [İzleme](#izleme)
-
-## Hızlı Başlangıç
-
-### 1. Bağımlılıkları Yükleyin
-
-```bash
-# ASGI sunucusu için uvicorn yükleyin
-pip install uvicorn
-
-# Veya tüm bağımlılıklarla birlikte yükleyin
-pip install -e .
-pip install uvicorn
-```
-
-### 2. Sunucuyu Çalıştırın
-
-```bash
-# Temel başlatma
-python run_asgi.py
-
-# Veya doğrudan uvicorn ile
-uvicorn asgi_app:app --host 0.0.0.0 --port 8000
-```
-
-Sunucu şu adreslerde kullanılabilir olacak:
-- MCP Endpoint: `http://localhost:8000/mcp/`
-- Sağlık Kontrolü: `http://localhost:8000/health`
-- API Durumu: `http://localhost:8000/status`
-
-## Yerel Geliştirme
-
-### Otomatik Yeniden Yükleme ile Geliştirme Sunucusu
-
-```bash
-python run_asgi.py --reload --log-level debug
-```
-
-### FastAPI Entegrasyonunu Kullanma
-
-Ek REST API endpoint'leri için:
-
-```bash
-uvicorn fastapi_app:app --reload
-```
-
-Bu şunları sağlar:
-- `/docs` adresinde interaktif API dokümantasyonu
-- `/api/tools` adresinde araç listesi
-- `/api/databases` adresinde veritabanı bilgileri
-
-### Ortam Değişkenleri
-
-`.env.example` dosyasını temel alarak bir `.env` dosyası oluşturun:
-
-```bash
-cp .env.example .env
-```
-
-Temel değişkenler:
-- `HOST`: Sunucu host adresi (varsayılan: 127.0.0.1)
-- `PORT`: Sunucu portu (varsayılan: 8000)
-- `ALLOWED_ORIGINS`: CORS kökenleri (virgülle ayrılmış)
-- `LOG_LEVEL`: Log seviyesi (debug, info, warning, error)
-
-## Production Dağıtımı
-
-### 1. Uvicorn ile Çoklu Worker Kullanımı
-
-```bash
-python run_asgi.py --host 0.0.0.0 --port 8000 --workers 4
-```
-
-### 2. Gunicorn Kullanımı
-
-```bash
-pip install gunicorn
-gunicorn asgi_app:app -w 4 -k uvicorn.workers.UvicornWorker --bind 0.0.0.0:8000
-```
-
-### 3. Nginx Reverse Proxy ile
-
-1. Nginx'i yükleyin
-2. Sağlanan `nginx.conf` dosyasını kullanın:
-
-```bash
-sudo cp nginx.conf /etc/nginx/sites-available/yargi-mcp
-sudo ln -s /etc/nginx/sites-available/yargi-mcp /etc/nginx/sites-enabled/
-sudo nginx -t
-sudo systemctl reload nginx
-```
-
-### 4. Systemd Servisi
-
-`/etc/systemd/system/yargi-mcp.service` dosyasını oluşturun:
-
-```ini
-[Unit]
-Description=Yargı MCP Server
-After=network.target
-
-[Service]
-Type=exec
-User=www-data
-WorkingDirectory=/opt/yargi-mcp
-Environment="PATH=/opt/yargi-mcp/venv/bin"
-ExecStart=/opt/yargi-mcp/venv/bin/uvicorn asgi_app:app --host 0.0.0.0 --port 8000 --workers 4
-Restart=on-failure
-RestartSec=5
-
-[Install]
-WantedBy=multi-user.target
-```
-
-Etkinleştirin ve başlatın:
-
-```bash
-sudo systemctl enable yargi-mcp
-sudo systemctl start yargi-mcp
-```
-
-## Cloud Dağıtımı
-
-### Heroku
-
-1. `Procfile` oluşturun:
-```
-web: uvicorn asgi_app:app --host 0.0.0.0 --port $PORT
-```
-
-2. Dağıtın:
-```bash
-heroku create uygulama-isminiz
-git push heroku main
-```
-
-### Railway
-
-1. `railway.json` ekleyin:
-```json
-{
-  "build": {
-    "builder": "NIXPACKS"
-  },
-  "deploy": {
-    "startCommand": "uvicorn asgi_app:app --host 0.0.0.0 --port $PORT"
-  }
-}
-```
-
-2. Railway CLI veya GitHub entegrasyonu ile dağıtın
-
-### Google Cloud Run
-
-1. Container oluşturun:
-```bash
-docker build -t yargi-mcp .
-docker tag yargi-mcp gcr.io/PROJE_ADINIZ/yargi-mcp
-docker push gcr.io/PROJE_ADINIZ/yargi-mcp
-```
-
-2. Dağıtın:
-```bash
-gcloud run deploy yargi-mcp \
-  --image gcr.io/PROJE_ADINIZ/yargi-mcp \
-  --platform managed \
-  --region us-central1 \
-  --allow-unauthenticated
-```
-
-### AWS Lambda (Mangum kullanarak)
-
-1. Mangum'u yükleyin:
-```bash
-pip install mangum
-```
-
-2. `lambda_handler.py` oluşturun:
-```python
-from mangum import Mangum
-from asgi_app import app
-
-handler = Mangum(app, lifespan="off")
-```
-
-3. AWS SAM veya Serverless Framework kullanarak dağıtın
-
-## Docker Dağıtımı
-
-### Tek Container
-
-```bash
-# Oluşturun
-docker build -t yargi-mcp .
-
-# Çalıştırın
-docker run -p 8000:8000 --env-file .env yargi-mcp
-```
-
-### Docker Compose
-
-```bash
-# Geliştirme
-docker-compose up
-
-# Nginx ile Production
-docker-compose --profile production up
-
-# Redis önbellekleme ile
-docker-compose --profile with-cache up
-```
-
-### Kubernetes
-
-Deployment YAML oluşturun:
-
-```yaml
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: yargi-mcp
-spec:
-  replicas: 3
-  selector:
-    matchLabels:
-      app: yargi-mcp
-  template:
-    metadata:
-      labels:
-        app: yargi-mcp
-    spec:
-      containers:
-      - name: yargi-mcp
-        image: yargi-mcp:latest
-        ports:
-        - containerPort: 8000
-        env:
-        - name: HOST
-          value: "0.0.0.0"
-        - name: PORT
-          value: "8000"
-        livenessProbe:
-          httpGet:
-            path: /health
-            port: 8000
-          initialDelaySeconds: 10
-          periodSeconds: 30
----
-apiVersion: v1
-kind: Service
-metadata:
-  name: yargi-mcp-service
-spec:
-  selector:
-    app: yargi-mcp
-  ports:
-  - port: 80
-    targetPort: 8000
-  type: LoadBalancer
-```
-
-## Güvenlik Hususları
-
-### 1. Kimlik Doğrulama
-
-`API_TOKEN` ortam değişkenini ayarlayarak token kimlik doğrulamasını etkinleştirin:
-
-```bash
-export API_TOKEN=gizli-token-degeri
-```
-
-Ardından isteklere ekleyin:
-```bash
-curl -H "Authorization: Bearer gizli-token-degeri" http://localhost:8000/api/tools
-```
-
-### 2. HTTPS/SSL
-
-Production için her zaman HTTPS kullanın:
-
-1. SSL sertifikası edinin (Let's Encrypt vb.)
-2. Nginx veya cloud sağlayıcıda yapılandırın
-3. `ALLOWED_ORIGINS` değerini https:// kullanacak şekilde güncelleyin
-
-### 3. Rate Limiting (Hız Sınırlama)
-
-Sağlanan Nginx yapılandırması rate limiting içerir:
-- API endpoint'leri: 10 istek/saniye
-- MCP endpoint: 100 istek/saniye
-
-### 4. CORS Yapılandırması
-
-Production için belirli kaynaklara izin verin:
-
-```bash
-ALLOWED_ORIGINS=https://app.sizindomain.com,https://www.sizindomain.com
-```
-
-## İzleme
-
-### Sağlık Kontrolleri
-
-`/health` endpoint'ini izleyin:
-
-```bash
-curl http://localhost:8000/health
-```
-
-Yanıt:
-```json
-{
-  "status": "healthy",
-  "timestamp": "2024-12-26T10:00:00",
-  "uptime_seconds": 3600,
-  "tools_operational": true
-}
-```
-
-### Loglama
-
-Ortam değişkeni ile log seviyesini yapılandırın:
-
-```bash
-LOG_LEVEL=info  # veya debug, warning, error
-```
-
-Loglar şuraya yazılır:
-- Konsol (stdout)
-- `logs/mcp_server.log` dosyası
-
-### Metrikler (Opsiyonel)
-
-OpenTelemetry desteği için:
-
-```bash
-pip install opentelemetry-instrumentation-fastapi
-```
-
-Ortam değişkenlerini ayarlayın:
-```bash
-OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
-OTEL_SERVICE_NAME=yargi-mcp-server
-```
-
-## Sorun Giderme
-
-### Port Zaten Kullanımda
-
-```bash
-# 8000 portunu kullanan işlemi bulun
-lsof -i :8000
-
-# İşlemi sonlandırın
-kill -9 <PID>
-```
-
-### İzin Hataları
-
-Dosya izinlerinin doğru olduğundan emin olun:
-
-```bash
-chmod +x run_asgi.py
-chown -R www-data:www-data /opt/yargi-mcp
-```
-
-### Bellek Sorunları
-
-Büyük belge işleme için worker belleğini artırın:
-
-```bash
-# systemd servisinde
-Environment="PYTHONMALLOC=malloc"
-LimitNOFILE=65536
-```
-
-### Zaman Aşımı Sorunları
-
-Zaman aşımlarını ayarlayın:
-1. Uvicorn: `--timeout-keep-alive 75`
-2. Nginx: `proxy_read_timeout 300s;`
-3. Cloud sağlayıcılar: Platform özel zaman aşımı ayarlarını kontrol edin
-
-## Performans Ayarlama
-
-### 1. Worker İşlemleri
-
-- Geliştirme: 1 worker
-- Production: CPU çekirdeği başına 2-4 worker
-
-### 2. Bağlantı Havuzlama
-
-Sunucu varsayılan olarak httpx ile bağlantı havuzlama kullanır.
-
-### 3. Önbellekleme (Gelecek Geliştirme)
-
-Redis önbellekleme docker-compose ile etkinleştirilebilir:
-
-```bash
-docker-compose --profile with-cache up
-```
-
-### 4. Veritabanı Zaman Aşımları
-
-`.env` dosyasında veritabanı başına zaman aşımlarını ayarlayın:
-
-```bash
-YARGITAY_TIMEOUT=60
-DANISTAY_TIMEOUT=60
-ANAYASA_TIMEOUT=90
-```
-
-## Destek
-
-Sorunlar ve sorular için:
-- GitHub Issues: https://github.com/saidsurucu/yargi-mcp/issues
-- Dokümantasyon: README.md dosyasına bakın

emsal_mcp_module/client.py

@@ -1,5 +1,6 @@
 # emsal_mcp_module/client.py
 
+import asyncio
 import httpx
 # from bs4 import BeautifulSoup # Uncomment if needed for advanced HTML pre-processing
 from typing import Dict, Any, List, Optional
@@ -153,7 +154,7 @@ class EmsalApiClient:
                 logger.warning(f"EmsalApiClient: Received empty or non-string HTML in 'data' field for Emsal ID {id}.")
                 return EmsalDocumentMarkdown(id=id, markdown_content=None, source_url=source_url)
 
-            markdown_content = self._clean_html_and_convert_to_markdown_emsal(html_content_from_api)
+            markdown_content = await asyncio.to_thread(self._clean_html_and_convert_to_markdown_emsal, html_content_from_api)
 
             return EmsalDocumentMarkdown(
                 id=id,

emsal_mcp_module/models.py

@@ -84,7 +84,7 @@ class EmsalApiResponseInnerData(BaseModel):
 
 class EmsalApiResponse(BaseModel):
     """Model for the complete search response from the Emsal API."""
-    data: EmsalApiResponseInnerData
+    data: Optional[EmsalApiResponseInnerData] = None
     metadata: Optional[Dict[str, Any]] = Field(None, description="Optional metadata (Meta Veri) from API, if any.")
 
 class EmsalDocumentMarkdown(BaseModel):

fly.toml

@@ -1,34 +0,0 @@
-# fly.toml app configuration file generated for yargi-mcp on 2025-06-29T00:23:47+03:00
-#
-# See https://fly.io/docs/reference/configuration/ for information about how to use this file.
-#
-
-app = 'yargi-mcp'
-primary_region = 'fra'
-
-[env]
-ENABLE_AUTH = "true"
-HOST = "0.0.0.0"
-PORT = "8000"
-LOG_LEVEL = "info"
-
-[build]
-
-[http_service]
-  internal_port = 8000
-  force_https = true
-  auto_stop_machines = 'off'
-  auto_start_machines = true
-  min_machines_running = 1
-  processes = ['app']
-
-[[vm]]
-  memory = '1gb'
-  cpu_kind = 'shared'
-  cpus = 1
-
-[checks.http_health]             # keep MCP /health live
-  type     = "http"
-  interval = "30s"
-  timeout  = "10s"
-  path     = "/health"

gib_mcp_module/__init__.py

@@ -0,0 +1 @@
+# gib_mcp_module/__init__.py

gib_mcp_module/client.py

@@ -0,0 +1,355 @@
+# gib_mcp_module/client.py
+
+import asyncio
+import httpx
+import io
+import logging
+import math
+from typing import Optional, Any, Dict
+from markitdown import MarkItDown
+
+from .models import (
+    GibSearchRequest,
+    GibOzelgeSummary,
+    GibSearchResult,
+    GibDocumentMarkdown,
+)
+
+logger = logging.getLogger(__name__)
+if not logger.hasHandlers():
+    logging.basicConfig(
+        level=logging.INFO,
+        format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
+    )
+
+
+class GibApiClient:
+    """
+    API client for searching and retrieving GİB özelgeler (Turkish Revenue
+    Administration tax rulings) via the public gib.gov.tr JSON API.
+
+    The endpoint is a single POST list endpoint; document retrieval is done
+    by filtering the same endpoint with an exact `id`.
+    """
+
+    BASE_URL = "https://gib.gov.tr/api"
+    LIST_PATH = "/gibportal/mevzuat/ozelge/list"
+    DOCUMENT_MARKDOWN_CHUNK_SIZE = 5000
+
+    # Fixed filter values required by the backend
+    _REQUIRED_STATUS = 2
+    _REQUIRED_DELETED = False
+    _REQUIRED_KTYPE = 99  # ktype=99 selects özelge
+    _SORT_FIELD = "ozelgeTarih"
+    _SORT_TYPE = "DESC"
+
+    def __init__(self, request_timeout: float = 60.0):
+        self.http_client = httpx.AsyncClient(
+            base_url=self.BASE_URL,
+            headers={
+                "Accept": "application/json",
+                "Accept-Language": "tr-TR,tr;q=0.9,en;q=0.7",
+                "Content-Type": "application/json",
+                "User-Agent": "Mozilla/5.0 (compatible; yargi-mcp/1.0; +https://github.com/saidsurucu/yargi-mcp)",
+            },
+            timeout=request_timeout,
+            verify=True,
+            follow_redirects=True,
+        )
+
+    @staticmethod
+    def _normalize_date(value: str, end_of_day: bool = False) -> Optional[str]:
+        """
+        Accept 'YYYY-MM-DD' or full ISO 8601; always return full ISO 8601.
+
+        GİB backend rejects date-only strings.
+        """
+        if not value:
+            return None
+        v = value.strip()
+        if not v:
+            return None
+        # Already ISO with time component
+        if "T" in v:
+            return v
+        # Simple YYYY-MM-DD - expand to start/end of day
+        suffix = "T23:59:59.999Z" if end_of_day else "T00:00:00.000Z"
+        return f"{v}{suffix}"
+
+    def _build_search_body(self, params: GibSearchRequest) -> Dict[str, Any]:
+        body: Dict[str, Any] = {
+            "status": self._REQUIRED_STATUS,
+            "deleted": self._REQUIRED_DELETED,
+            "ktype": self._REQUIRED_KTYPE,
+        }
+
+        keywords = params.keywords.strip()
+        kanun_no = params.kanunNo.strip()
+        # Frontend sets title/kanunNo/description to the SAME value; the backend
+        # ORs across them. If the caller supplies both, combine them so kanun_no
+        # still biases toward ruling text, while keywords remain primary.
+        search_term = keywords or kanun_no
+        if keywords and kanun_no and kanun_no not in keywords:
+            search_term = f"{keywords} {kanun_no}"
+        if search_term:
+            body["title"] = search_term
+            body["kanunNo"] = search_term
+            body["description"] = search_term
+
+        if params.ozelgeNo.strip():
+            body["ozelgeNo"] = params.ozelgeNo.strip()
+
+        if params.kanunId and params.kanunId > 0:
+            body["kanunIds"] = [params.kanunId]
+
+        start_iso = self._normalize_date(params.ozelgeStartDate, end_of_day=False)
+        end_iso = self._normalize_date(params.ozelgeEndDate, end_of_day=True)
+        if start_iso:
+            body["ozelgeStartDate"] = start_iso
+        if end_iso:
+            body["ozelgeEndDate"] = end_iso
+
+        return body
+
+    def _build_query_params(self, page_1_indexed: int, page_size: int) -> Dict[str, Any]:
+        # API expects 0-indexed page
+        zero_indexed = max(0, page_1_indexed - 1)
+        return {
+            "page": zero_indexed,
+            "size": page_size,
+            "sortFieldName": self._SORT_FIELD,
+            "sortType": self._SORT_TYPE,
+        }
+
+    @staticmethod
+    def _to_summary(item: Dict[str, Any]) -> Optional[GibOzelgeSummary]:
+        if not isinstance(item, dict):
+            return None
+        raw_id = item.get("id")
+        if raw_id is None:
+            return None
+        try:
+            ozelge_id = int(raw_id)
+        except (TypeError, ValueError):
+            return None
+        return GibOzelgeSummary(
+            id=ozelge_id,
+            ozelgeNo=item.get("ozelgeNo"),
+            ozelgeTarih=item.get("ozelgeTarih"),
+            title=item.get("title"),
+            kanunNo=item.get("kanunNo"),
+            kanunTitle=item.get("kanunTitle"),
+            siteLink=item.get("siteLink"),
+        )
+
+    async def search_ozelge(self, params: GibSearchRequest) -> GibSearchResult:
+        """Search GİB özelgeler."""
+        body = self._build_search_body(params)
+        query = self._build_query_params(params.page, params.pageSize)
+        logger.info(
+            "GibApiClient: search page=%s size=%s body_keys=%s",
+            params.page, params.pageSize, sorted(body.keys()),
+        )
+
+        try:
+            resp = await self.http_client.post(self.LIST_PATH, params=query, json=body)
+            resp.raise_for_status()
+            payload = resp.json()
+        except httpx.HTTPStatusError as e:
+            logger.error("GibApiClient: HTTP %s during search", e.response.status_code)
+            return GibSearchResult(
+                ozelgeler=[],
+                total_results=0,
+                total_pages=0,
+                current_page=params.page,
+                page_size=params.pageSize,
+            )
+        except Exception as e:
+            logger.error("GibApiClient: search request failed: %s", e)
+            return GibSearchResult(
+                ozelgeler=[],
+                total_results=0,
+                total_pages=0,
+                current_page=params.page,
+                page_size=params.pageSize,
+            )
+
+        container = (payload or {}).get("resultContainer") or {}
+        raw_items = container.get("content") or []
+
+        summaries = []
+        for raw in raw_items:
+            summary = self._to_summary(raw)
+            if summary is not None:
+                summaries.append(summary)
+
+        total_results = container.get("totalElements") or 0
+        total_pages = container.get("totalPages") or 0
+        try:
+            total_results = int(total_results)
+        except (TypeError, ValueError):
+            total_results = 0
+        try:
+            total_pages = int(total_pages)
+        except (TypeError, ValueError):
+            total_pages = 0
+
+        return GibSearchResult(
+            ozelgeler=summaries,
+            total_results=total_results,
+            total_pages=total_pages,
+            current_page=params.page,
+            page_size=params.pageSize,
+        )
+
+    def _convert_html_to_markdown(self, html_content: str) -> Optional[str]:
+        """Convert HTML content to Markdown using MarkItDown with BytesIO."""
+        if not html_content:
+            return None
+        try:
+            html_bytes = html_content.encode("utf-8")
+            html_stream = io.BytesIO(html_bytes)
+            md_converter = MarkItDown(enable_plugins=False)
+            result = md_converter.convert(html_stream)
+            return result.text_content
+        except Exception as e:
+            logger.error("GibApiClient: HTML→Markdown conversion failed: %s", e)
+            return None
+
+    @staticmethod
+    def _build_header_block(item: Dict[str, Any]) -> str:
+        """Build a small Markdown header block summarising the ruling metadata."""
+        parts = []
+        title = item.get("title")
+        if title:
+            parts.append(f"# {title}")
+        meta_lines = []
+        if item.get("ozelgeNo"):
+            meta_lines.append(f"**Sayı:** {item['ozelgeNo']}")
+        if item.get("ozelgeTarih"):
+            meta_lines.append(f"**Tarih:** {item['ozelgeTarih']}")
+        if item.get("kanunTitle"):
+            kanun_no = item.get("kanunNo")
+            if kanun_no:
+                meta_lines.append(f"**Kanun:** {item['kanunTitle']} ({kanun_no})")
+            else:
+                meta_lines.append(f"**Kanun:** {item['kanunTitle']}")
+        if item.get("siteLink"):
+            meta_lines.append(f"**Kaynak:** {item['siteLink']}")
+        if meta_lines:
+            parts.append("\n".join(meta_lines))
+        return "\n\n".join(parts).strip()
+
+    async def get_ozelge_document(
+        self, ozelge_id: int, page_number: int = 1
+    ) -> GibDocumentMarkdown:
+        """Retrieve a single özelge and return its paginated Markdown form."""
+        logger.info(
+            "GibApiClient: fetching özelge id=%s page=%s", ozelge_id, page_number
+        )
+
+        if not isinstance(ozelge_id, int) or ozelge_id <= 0:
+            return GibDocumentMarkdown(
+                ozelge_id=ozelge_id if isinstance(ozelge_id, int) else 0,
+                current_page=page_number,
+                total_pages=0,
+                is_paginated=False,
+                error_message="ozelge_id must be a positive integer",
+            )
+
+        body = {
+            "status": self._REQUIRED_STATUS,
+            "deleted": self._REQUIRED_DELETED,
+            "ktype": self._REQUIRED_KTYPE,
+            "id": ozelge_id,
+        }
+        query = {"page": 0, "size": 1}
+
+        try:
+            resp = await self.http_client.post(self.LIST_PATH, params=query, json=body)
+            resp.raise_for_status()
+            payload = resp.json()
+        except httpx.HTTPStatusError as e:
+            msg = f"HTTP {e.response.status_code} when fetching özelge {ozelge_id}"
+            logger.error("GibApiClient: %s", msg)
+            return GibDocumentMarkdown(
+                ozelge_id=ozelge_id,
+                current_page=page_number,
+                total_pages=0,
+                is_paginated=False,
+                error_message=msg,
+            )
+        except Exception as e:
+            msg = f"Request failed: {e}"
+            logger.error("GibApiClient: %s", msg)
+            return GibDocumentMarkdown(
+                ozelge_id=ozelge_id,
+                current_page=page_number,
+                total_pages=0,
+                is_paginated=False,
+                error_message=msg,
+            )
+
+        container = (payload or {}).get("resultContainer") or {}
+        content = container.get("content") or []
+        if not content:
+            return GibDocumentMarkdown(
+                ozelge_id=ozelge_id,
+                current_page=page_number,
+                total_pages=0,
+                is_paginated=False,
+                error_message=f"Özelge {ozelge_id} not found",
+            )
+
+        item = content[0] if isinstance(content[0], dict) else {}
+        description_html = item.get("description") or ""
+        markdown_body = (await asyncio.to_thread(self._convert_html_to_markdown, description_html)) or ""
+        header_block = self._build_header_block(item)
+
+        if header_block and markdown_body:
+            full_markdown = f"{header_block}\n\n---\n\n{markdown_body}"
+        else:
+            full_markdown = header_block or markdown_body
+
+        if not full_markdown.strip():
+            return GibDocumentMarkdown(
+                ozelge_id=ozelge_id,
+                ozelge_no=item.get("ozelgeNo"),
+                title=item.get("title"),
+                ozelge_tarih=item.get("ozelgeTarih"),
+                kanun_title=item.get("kanunTitle"),
+                kanun_no=item.get("kanunNo"),
+                site_link=item.get("siteLink"),
+                current_page=page_number,
+                total_pages=0,
+                is_paginated=False,
+                error_message="Document body is empty",
+            )
+
+        total_pages = max(
+            1, math.ceil(len(full_markdown) / self.DOCUMENT_MARKDOWN_CHUNK_SIZE)
+        )
+        current_page_clamped = max(1, min(page_number, total_pages))
+        start = (current_page_clamped - 1) * self.DOCUMENT_MARKDOWN_CHUNK_SIZE
+        end = start + self.DOCUMENT_MARKDOWN_CHUNK_SIZE
+        chunk = full_markdown[start:end]
+
+        return GibDocumentMarkdown(
+            ozelge_id=ozelge_id,
+            ozelge_no=item.get("ozelgeNo"),
+            title=item.get("title"),
+            ozelge_tarih=item.get("ozelgeTarih"),
+            kanun_title=item.get("kanunTitle"),
+            kanun_no=item.get("kanunNo"),
+            site_link=item.get("siteLink"),
+            markdown_chunk=chunk,
+            current_page=current_page_clamped,
+            total_pages=total_pages,
+            is_paginated=total_pages > 1,
+            error_message=None,
+        )
+
+    async def close_client_session(self):
+        if hasattr(self, "http_client") and self.http_client and not self.http_client.is_closed:
+            await self.http_client.aclose()
+            logger.info("GibApiClient: HTTP client session closed.")

gib_mcp_module/models.py

@@ -0,0 +1,64 @@
+# gib_mcp_module/models.py
+
+from pydantic import BaseModel, Field
+from typing import List, Optional
+
+
+class GibSearchRequest(BaseModel):
+    """
+    Request model for searching GİB özelgeler (Turkish Revenue Administration tax rulings).
+
+    GİB (Gelir İdaresi Başkanlığı) publishes official tax-ruling letters
+    ("özelge") responding to taxpayer questions on VAT, income tax,
+    corporate tax, stamp duty, and other tax matters. 18,000+ rulings
+    are searchable via the public gib.gov.tr API.
+    """
+    keywords: str = Field("", description="Keywords searched across title, kanunNo and description (Turkish)")
+    ozelgeNo: str = Field("", description="Exact özelge reference number (e.g., 'E-40247694-130-15524')")
+    kanunNo: str = Field("", description="Law number filter, e.g. '3065' for KDV")
+    kanunId: int = Field(0, description="Optional numeric law ID filter (0=ignore)")
+    ozelgeStartDate: str = Field("", description="Start date YYYY-MM-DD or full ISO 8601")
+    ozelgeEndDate: str = Field("", description="End date YYYY-MM-DD or full ISO 8601")
+    page: int = Field(1, ge=1, description="Page number (1-indexed)")
+    pageSize: int = Field(10, ge=1, le=50, description="Results per page (1-50)")
+
+
+class GibOzelgeSummary(BaseModel):
+    """Summary of a single GİB özelge from search results (no full HTML)."""
+    id: int = Field(..., description="Numeric özelge ID for document retrieval")
+    ozelgeNo: Optional[str] = Field(None, description="Official ruling reference number")
+    ozelgeTarih: Optional[str] = Field(None, description="Ruling date (ISO datetime)")
+    title: Optional[str] = Field(None, description="Subject/title of the ruling")
+    kanunNo: Optional[str] = Field(None, description="Law number (e.g., '3065')")
+    kanunTitle: Optional[str] = Field(None, description="Law title (e.g., 'KATMA DEĞER VERGİSİ KANUNU')")
+    siteLink: Optional[str] = Field(None, description="Direct URL to the ruling on gib.gov.tr")
+
+
+class GibSearchResult(BaseModel):
+    """Response model for GİB özelge search results."""
+    ozelgeler: List[GibOzelgeSummary] = Field(default_factory=list, description="Matching özelge summaries")
+    total_results: int = Field(0, description="Total number of matching özelgeler across all pages")
+    total_pages: int = Field(0, description="Total number of pages for this query")
+    current_page: int = Field(1, description="Current page (1-indexed)")
+    page_size: int = Field(10, description="Results per page")
+
+
+class GibDocumentMarkdown(BaseModel):
+    """
+    GİB özelge document converted to paginated Markdown.
+
+    Long rulings are split into 5000-character chunks; request successive
+    pages via page_number to read the full text.
+    """
+    ozelge_id: int = Field(..., description="Numeric özelge ID")
+    ozelge_no: Optional[str] = Field(None, description="Official ruling reference number")
+    title: Optional[str] = Field(None, description="Subject/title of the ruling")
+    ozelge_tarih: Optional[str] = Field(None, description="Ruling date (ISO datetime)")
+    kanun_title: Optional[str] = Field(None, description="Related law title")
+    kanun_no: Optional[str] = Field(None, description="Related law number")
+    site_link: Optional[str] = Field(None, description="Direct URL to the ruling on gib.gov.tr")
+    markdown_chunk: Optional[str] = Field(None, description="Current 5000-character Markdown chunk")
+    current_page: int = Field(1, description="Current page number (1-indexed)")
+    total_pages: int = Field(0, description="Total pages for the full Markdown content")
+    is_paginated: bool = Field(False, description="True if split across multiple pages")
+    error_message: Optional[str] = Field(None, description="Populated when retrieval failed")

kik_mcp_module/client_v2.py

@@ -0,0 +1,507 @@
+# kik_mcp_module/client_v2.py
+
+import asyncio
+import base64
+import httpx
+import logging
+import uuid
+import ssl
+import os
+from typing import Optional
+from datetime import datetime
+
+# Cryptography imports for AES-256-CBC encryption of document IDs
+try:
+    from cryptography.hazmat.primitives.ciphers import Cipher, algorithms, modes
+    from cryptography.hazmat.backends import default_backend
+    HAS_CRYPTOGRAPHY = True
+except ImportError:
+    HAS_CRYPTOGRAPHY = False
+
+from .models_v2 import (
+    KikV2DecisionType, KikV2SearchPayload, KikV2SearchPayloadDk, KikV2SearchPayloadMk,
+    KikV2RequestData, KikV2QueryRequest, KikV2KeyValuePair, 
+    KikV2SearchResponse, KikV2SearchResponseDk, KikV2SearchResponseMk,
+    KikV2SearchResult, KikV2CompactDecision, KikV2DocumentMarkdown
+)
+
+logger = logging.getLogger(__name__)
+
+class KikV2ApiClient:
+    """
+    New KIK v2 API Client for https://ekapv2.kik.gov.tr
+    
+    This client uses the modern JSON-based API endpoint that provides
+    better structured data compared to the legacy form-based API.
+    """
+    
+    BASE_URL = "https://ekapv2.kik.gov.tr"
+    
+    # Endpoint mappings for different decision types
+    ENDPOINTS = {
+        KikV2DecisionType.UYUSMAZLIK: "/b_ihalearaclari/api/KurulKararlari/GetKurulKararlari",
+        KikV2DecisionType.DUZENLEYICI: "/b_ihalearaclari/api/KurulKararlari/GetKurulKararlariDk",
+        KikV2DecisionType.MAHKEME: "/b_ihalearaclari/api/KurulKararlari/GetKurulKararlariMk"
+    }
+
+    # AES-256-CBC encryption key for document ID encryption (reverse engineered from ekapv2.kik.gov.tr Angular app)
+    # This key is used to encrypt numeric gundemMaddesiId values to 64-character hex hashes for document URLs
+    DOCUMENT_ID_ENCRYPTION_KEY = bytes([
+        236, 193, 164, 43, 12, 135, 121, 170, 4, 244, 123, 219, 82, 158, 124, 174,
+        174, 228, 219, 174, 208, 104, 174, 120, 32, 76, 250, 4, 143, 159, 211, 176
+    ])
+
+    # AES-192-CBC key (environment.r8fact) used by the Angular HTTP interceptor to sign every
+    # request. The server decrypts X-Custom-Request-Ts and rejects stale timestamps with
+    # HTTP 401 "İstek zaman aşımına uğradı.", so these headers MUST be generated per-request
+    # with the current timestamp (see _generate_security_headers).
+    REQUEST_SIGNING_KEY = b"Qm2LtXR0aByP69vZNKef4wMJ"  # UTF-8 bytes, 24 chars -> AES-192
+
+    @staticmethod
+    def encrypt_document_id(numeric_id: str) -> str:
+        """
+        Encrypt a numeric KİK gundemMaddesiId to the 64-character hex hash
+        used in document URLs.
+
+        Algorithm: AES-256-CBC with PKCS7 padding
+        Output format: IV (16 bytes hex) + Ciphertext (16 bytes hex) = 64 chars
+
+        Args:
+            numeric_id: The numeric document ID from search results (e.g., "177280")
+
+        Returns:
+            64-character hex string for use in document URL KararId parameter
+        """
+        if not HAS_CRYPTOGRAPHY:
+            raise ImportError("cryptography library required for document ID encryption")
+
+        # Generate random IV (16 bytes)
+        iv = os.urandom(16)
+
+        # Create AES-CBC cipher with the encryption key
+        cipher = Cipher(
+            algorithms.AES(KikV2ApiClient.DOCUMENT_ID_ENCRYPTION_KEY),
+            modes.CBC(iv),
+            backend=default_backend()
+        )
+        encryptor = cipher.encryptor()
+
+        # Encode plaintext and apply PKCS7 padding
+        plaintext = numeric_id.encode('utf-8')
+        block_size = 16
+        padding_len = block_size - (len(plaintext) % block_size)
+        padded_plaintext = plaintext + bytes([padding_len] * padding_len)
+
+        # Encrypt
+        ciphertext = encryptor.update(padded_plaintext) + encryptor.finalize()
+
+        # Return IV + ciphertext as lowercase hex (64 characters total)
+        return iv.hex() + ciphertext.hex()
+
+    def __init__(self, request_timeout: float = 60.0):
+        # Create SSL context with legacy server support
+        ssl_context = ssl.create_default_context()
+        ssl_context.check_hostname = False
+        ssl_context.verify_mode = ssl.CERT_NONE
+        
+        # Enable legacy server connect option for older SSL implementations (Python 3.12+)
+        if hasattr(ssl, 'OP_LEGACY_SERVER_CONNECT'):
+            ssl_context.options |= ssl.OP_LEGACY_SERVER_CONNECT
+        
+        # Set broader cipher suite support including legacy ciphers
+        ssl_context.set_ciphers('ALL:!aNULL:!eNULL:!EXPORT:!DES:!RC4:!MD5:!PSK:!SRP:!CAMELLIA')
+        
+        self.http_client = httpx.AsyncClient(
+            base_url=self.BASE_URL,
+            verify=ssl_context,
+            headers={
+                "Accept": "application/json",
+                "Accept-Language": "tr",
+                "Content-Type": "application/json",
+                "Origin": self.BASE_URL,
+                "Referer": f"{self.BASE_URL}/sorgulamalar/kurul-kararlari",
+                "Sec-Fetch-Dest": "empty",
+                "Sec-Fetch-Mode": "cors", 
+                "Sec-Fetch-Site": "same-origin",
+                "User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/139.0.0.0 Safari/537.36",
+                "api-version": "v1",
+                "sec-ch-ua": '"Not;A=Brand";v="99", "Google Chrome";v="139", "Chromium";v="139"',
+                "sec-ch-ua-mobile": "?0",
+                "sec-ch-ua-platform": '"macOS"'
+            },
+            timeout=request_timeout
+        )
+        
+        # Generate security headers (these might need to be updated based on API requirements)
+        self.security_headers = self._generate_security_headers()
+    
+    def _sign_request_value(self, plaintext: str, iv: bytes) -> str:
+        """AES-192-CBC encrypt a value with the request signing key, return base64 ciphertext."""
+        cipher = Cipher(
+            algorithms.AES(self.REQUEST_SIGNING_KEY),
+            modes.CBC(iv),
+            backend=default_backend()
+        )
+        encryptor = cipher.encryptor()
+        data = plaintext.encode("utf-8")
+        block_size = 16
+        padding_len = block_size - (len(data) % block_size)
+        padded = data + bytes([padding_len] * padding_len)
+        ciphertext = encryptor.update(padded) + encryptor.finalize()
+        return base64.b64encode(ciphertext).decode("ascii")
+
+    def _generate_security_headers(self) -> dict:
+        """
+        Generate the custom security headers required by the KIK v2 API.
+
+        Mirrors the Angular HTTP interceptor on ekapv2.kik.gov.tr: a random GUID and a
+        current-timestamp (epoch milliseconds) are AES-192-CBC encrypted with environment.r8fact
+        using a fresh random IV. The IV is sent as -Siv, the encrypted timestamp as -Ts, and the
+        encrypted GUID as -R8id. The server validates the decrypted timestamp's freshness, so these
+        MUST be regenerated on every request; stale values yield HTTP 401 "İstek zaman aşımına uğradı.".
+        """
+        if not HAS_CRYPTOGRAPHY:
+            raise ImportError("cryptography library required for KIK v2 request signing")
+
+        request_guid = str(uuid.uuid4())
+        iv = os.urandom(16)
+        timestamp_ms = str(int(datetime.now().timestamp() * 1000))
+
+        return {
+            "X-Custom-Request-Guid": request_guid,
+            "X-Custom-Request-R8id": self._sign_request_value(request_guid, iv),
+            "X-Custom-Request-Siv": base64.b64encode(iv).decode("ascii"),
+            "X-Custom-Request-Ts": self._sign_request_value(timestamp_ms, iv),
+        }
+    
+    def _build_search_payload(self, 
+                             decision_type: KikV2DecisionType,
+                             karar_metni: str = "",
+                             karar_no: str = "",
+                             basvuran: str = "",
+                             idare_adi: str = "",
+                             baslangic_tarihi: str = "",
+                             bitis_tarihi: str = ""):
+        """Build the search payload for KIK v2 API."""
+        
+        key_value_pairs = []
+        
+        # Add non-empty search criteria
+        if karar_metni:
+            key_value_pairs.append(KikV2KeyValuePair(key="KararMetni", value=karar_metni))
+        
+        if karar_no:
+            key_value_pairs.append(KikV2KeyValuePair(key="KararNo", value=karar_no))
+            
+        if basvuran:
+            key_value_pairs.append(KikV2KeyValuePair(key="BasvuranAdi", value=basvuran))
+            
+        if idare_adi:
+            key_value_pairs.append(KikV2KeyValuePair(key="IdareAdi", value=idare_adi))
+            
+        if baslangic_tarihi:
+            key_value_pairs.append(KikV2KeyValuePair(key="BaslangicTarihi", value=baslangic_tarihi))
+            
+        if bitis_tarihi:
+            key_value_pairs.append(KikV2KeyValuePair(key="BitisTarihi", value=bitis_tarihi))
+        
+        # If no search criteria provided, use a generic search
+        if not key_value_pairs:
+            key_value_pairs.append(KikV2KeyValuePair(key="KararMetni", value=""))
+        
+        query_request = KikV2QueryRequest(keyValueOfstringanyType=key_value_pairs)
+        request_data = KikV2RequestData(keyValuePairs=query_request)
+        
+        # Return appropriate payload based on decision type
+        if decision_type == KikV2DecisionType.UYUSMAZLIK:
+            return KikV2SearchPayload(sorgulaKurulKararlari=request_data)
+        elif decision_type == KikV2DecisionType.DUZENLEYICI:
+            return KikV2SearchPayloadDk(sorgulaKurulKararlariDk=request_data)
+        elif decision_type == KikV2DecisionType.MAHKEME:
+            return KikV2SearchPayloadMk(sorgulaKurulKararlariMk=request_data)
+        else:
+            raise ValueError(f"Unsupported decision type: {decision_type}")
+    
+    async def search_decisions(self,
+                              decision_type: KikV2DecisionType = KikV2DecisionType.UYUSMAZLIK,
+                              karar_metni: str = "",
+                              karar_no: str = "",
+                              basvuran: str = "",
+                              idare_adi: str = "",
+                              baslangic_tarihi: str = "",
+                              bitis_tarihi: str = "") -> KikV2SearchResult:
+        """
+        Search KIK decisions using the v2 API.
+        
+        Args:
+            decision_type: Type of decision to search (uyusmazlik/duzenleyici/mahkeme)
+            karar_metni: Decision text search
+            karar_no: Decision number (e.g., "2025/UH.II-1801")
+            basvuran: Applicant name
+            idare_adi: Administration name
+            baslangic_tarihi: Start date (YYYY-MM-DD format)
+            bitis_tarihi: End date (YYYY-MM-DD format)
+            
+        Returns:
+            KikV2SearchResult with compact decision list
+        """
+        
+        logger.info(f"KikV2ApiClient: Searching {decision_type.value} decisions with criteria - karar_metni: '{karar_metni}', karar_no: '{karar_no}', basvuran: '{basvuran}'")
+        
+        try:
+            # Build request payload
+            payload = self._build_search_payload(
+                decision_type=decision_type,
+                karar_metni=karar_metni,
+                karar_no=karar_no,
+                basvuran=basvuran,
+                idare_adi=idare_adi,
+                baslangic_tarihi=baslangic_tarihi,
+                bitis_tarihi=bitis_tarihi
+            )
+            
+            # Update security headers for this request
+            headers = {**self.http_client.headers, **self._generate_security_headers()}
+            
+            # Get the appropriate endpoint for this decision type
+            endpoint = self.ENDPOINTS[decision_type]
+            
+            # Make API request
+            response = await self.http_client.post(
+                endpoint,
+                json=payload.model_dump(),
+                headers=headers
+            )
+            
+            response.raise_for_status()
+            response_data = response.json()
+            
+            logger.debug(f"KikV2ApiClient: Raw API response structure: {type(response_data)}")
+            
+            # Parse the API response based on decision type
+            if decision_type == KikV2DecisionType.UYUSMAZLIK:
+                api_response = KikV2SearchResponse(**response_data)
+                result_data = api_response.SorgulaKurulKararlariResponse.SorgulaKurulKararlariResult
+            elif decision_type == KikV2DecisionType.DUZENLEYICI:
+                api_response = KikV2SearchResponseDk(**response_data)
+                result_data = api_response.SorgulaKurulKararlariDkResponse.SorgulaKurulKararlariDkResult
+            elif decision_type == KikV2DecisionType.MAHKEME:
+                api_response = KikV2SearchResponseMk(**response_data)
+                result_data = api_response.SorgulaKurulKararlariMkResponse.SorgulaKurulKararlariMkResult
+            else:
+                raise ValueError(f"Unsupported decision type: {decision_type}")
+                
+            # Check for API errors
+            if result_data.hataKodu and result_data.hataKodu != "0":
+                logger.warning(f"KikV2ApiClient: API returned error - Code: {result_data.hataKodu}, Message: {result_data.hataMesaji}")
+                return KikV2SearchResult(
+                    decisions=[],
+                    total_records=0,
+                    page=1,
+                    error_code=result_data.hataKodu,
+                    error_message=result_data.hataMesaji
+                )
+            
+            # Convert to compact format
+            compact_decisions = []
+            total_count = 0
+            
+            for decision_group in result_data.KurulKararTutanakDetayListesi:
+                for decision_detail in decision_group.KurulKararTutanakDetayi:
+                    compact_decision = KikV2CompactDecision(
+                        kararNo=decision_detail.kararNo,
+                        kararTarihi=decision_detail.kararTarihi,
+                        basvuran=decision_detail.basvuran,
+                        idareAdi=decision_detail.idareAdi,
+                        basvuruKonusu=decision_detail.basvuruKonusu,
+                        gundemMaddesiId=decision_detail.gundemMaddesiId,
+                        decision_type=decision_type.value
+                    )
+                    compact_decisions.append(compact_decision)
+                    total_count += 1
+            
+            logger.info(f"KikV2ApiClient: Found {total_count} decisions")
+            
+            return KikV2SearchResult(
+                decisions=compact_decisions,
+                total_records=total_count,
+                page=1,
+                error_code="0",
+                error_message=""
+            )
+            
+        except httpx.HTTPStatusError as e:
+            logger.error(f"KikV2ApiClient: HTTP error during search: {e.response.status_code} - {e.response.text}")
+            return KikV2SearchResult(
+                decisions=[],
+                total_records=0, 
+                page=1,
+                error_code="HTTP_ERROR",
+                error_message=f"HTTP {e.response.status_code}: {e.response.text}"
+            )
+        except Exception as e:
+            logger.error(f"KikV2ApiClient: Unexpected error during search: {str(e)}")
+            return KikV2SearchResult(
+                decisions=[],
+                total_records=0,
+                page=1, 
+                error_code="UNEXPECTED_ERROR",
+                error_message=str(e)
+            )
+    
+    async def get_document_markdown(self, document_id: str) -> KikV2DocumentMarkdown:
+        """
+        Get KİK decision document content in Markdown format.
+        
+        This method uses a two-step process:
+        1. Call GetSorgulamaUrl endpoint to get the actual document URL
+        2. Use httpx to fetch the document content
+        
+        Args:
+            document_id: The gundemMaddesiId from search results
+            
+        Returns:
+            KikV2DocumentMarkdown with document content converted to Markdown
+        """
+        
+        logger.info(f"KikV2ApiClient: Getting document for ID: {document_id}")
+        
+        if not document_id or not document_id.strip():
+            return KikV2DocumentMarkdown(
+                document_id=document_id,
+                kararNo="",
+                markdown_content="",
+                source_url="",
+                error_message="Document ID is required"
+            )
+        
+        try:
+            # Step 1: Get the actual document URL using GetSorgulamaUrl endpoint
+            logger.info(f"KikV2ApiClient: Step 1 - Getting document URL for ID: {document_id}")
+            
+            # Update security headers for this request
+            headers = {**self.http_client.headers, **self._generate_security_headers()}
+            
+            # Call GetSorgulamaUrl to get the real document URL
+            url_payload = {"sorguSayfaTipi": 2}  # As shown in curl example
+            
+            url_response = await self.http_client.post(
+                "/b_ihalearaclari/api/KurulKararlari/GetSorgulamaUrl",
+                json=url_payload,
+                headers=headers
+            )
+            
+            url_response.raise_for_status()
+            url_data = url_response.json()
+            
+            # Get the base document URL from API response
+            base_document_url = url_data.get("sorgulamaUrl", "")
+            if not base_document_url:
+                return KikV2DocumentMarkdown(
+                    document_id=document_id,
+                    kararNo="",
+                    markdown_content="",
+                    source_url="",
+                    error_message="Could not get document URL from GetSorgulamaUrl API"
+                )
+            
+            # If document_id is numeric, encrypt it to get the KararId hash
+            # The web interface uses AES-256-CBC encrypted hashes for document URLs
+            karar_id = document_id
+            if document_id.isdigit():
+                try:
+                    karar_id = self.encrypt_document_id(document_id)
+                    logger.info(f"KikV2ApiClient: Encrypted numeric ID {document_id} to hash: {karar_id}")
+                except Exception as enc_error:
+                    logger.warning(f"KikV2ApiClient: Could not encrypt document ID, using as-is: {enc_error}")
+
+            # Construct full document URL with the encrypted KararId
+            document_url = f"{base_document_url}?KararId={karar_id}"
+            logger.info(f"KikV2ApiClient: Step 2 - Retrieved document URL: {document_url}")
+
+        except Exception as e:
+            logger.error(f"KikV2ApiClient: Error getting document URL for ID {document_id}: {str(e)}")
+            # Fallback to old method if GetSorgulamaUrl fails
+            # Also encrypt numeric IDs in fallback path
+            karar_id = document_id
+            if document_id.isdigit():
+                try:
+                    karar_id = self.encrypt_document_id(document_id)
+                    logger.info(f"KikV2ApiClient: Encrypted numeric ID in fallback: {karar_id}")
+                except Exception as enc_error:
+                    logger.warning(f"KikV2ApiClient: Could not encrypt in fallback: {enc_error}")
+            document_url = f"https://ekap.kik.gov.tr/EKAP/Vatandas/KurulKararGoster.aspx?KararId={karar_id}"
+            logger.info(f"KikV2ApiClient: Falling back to direct URL: {document_url}")
+        
+        try:
+            # Step 2: Use httpx to get the document content
+            logger.info(f"KikV2ApiClient: Step 2 - Using httpx to retrieve document from: {document_url}")
+
+            # Create a separate httpx client for document retrieval with HTML headers
+            doc_ssl_context = ssl.create_default_context()
+            doc_ssl_context.check_hostname = False
+            doc_ssl_context.verify_mode = ssl.CERT_NONE
+            if hasattr(ssl, 'OP_LEGACY_SERVER_CONNECT'):
+                doc_ssl_context.options |= ssl.OP_LEGACY_SERVER_CONNECT
+            doc_ssl_context.set_ciphers('ALL:!aNULL:!eNULL:!EXPORT:!DES:!RC4:!MD5:!PSK:!SRP:!CAMELLIA')
+
+            async with httpx.AsyncClient(
+                verify=doc_ssl_context,
+                headers={
+                    "Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8",
+                    "Accept-Language": "tr,en-US;q=0.5",
+                    "User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/139.0.0.0 Safari/537.36"
+                },
+                timeout=60.0,
+                follow_redirects=True
+            ) as doc_client:
+                response = await doc_client.get(document_url)
+                response.raise_for_status()
+                html_content = response.text
+                logger.info(f"KikV2ApiClient: Retrieved content via httpx, length: {len(html_content)}")
+            
+            # Convert HTML to Markdown using MarkItDown with BytesIO
+            try:
+                from markitdown import MarkItDown
+                from io import BytesIO
+                
+                md = MarkItDown()
+                html_bytes = html_content.encode('utf-8')
+                html_stream = BytesIO(html_bytes)
+                
+                # markitdown is sync; offload to thread so HTML parsing doesn't
+                # block the event-loop / other in-flight MCP requests.
+                result = await asyncio.to_thread(md.convert_stream, html_stream, file_extension=".html")
+                markdown_content = result.text_content
+                
+                return KikV2DocumentMarkdown(
+                    document_id=document_id,
+                    kararNo="",
+                    markdown_content=markdown_content,
+                    source_url=document_url,
+                    error_message=""
+                )
+                
+            except ImportError:
+                return KikV2DocumentMarkdown(
+                    document_id=document_id,
+                    kararNo="",
+                    markdown_content="MarkItDown library not available",
+                    source_url=document_url,
+                    error_message="MarkItDown library not installed"
+                )
+                
+        except Exception as e:
+            logger.error(f"KikV2ApiClient: Error retrieving document {document_id}: {str(e)}")
+            return KikV2DocumentMarkdown(
+                document_id=document_id,
+                kararNo="",
+                markdown_content="",
+                source_url=document_url,
+                error_message=str(e)
+            )
+    
+    async def close_client_session(self):
+        """Close HTTP client session."""
+        await self.http_client.aclose()
+        logger.info("KikV2ApiClient: HTTP client session closed.")

kik_mcp_module/models.py

@@ -1,74 +0,0 @@
-# kik_mcp_module/models.py
-from pydantic import BaseModel, Field, HttpUrl, computed_field, ConfigDict
-from typing import List, Optional
-from enum import Enum
-import base64 # Base64 encoding/decoding için
-
-class KikKararTipi(str, Enum):
-    """Enum for KIK (Public Procurement Authority) Decision Types."""
-    UYUSMAZLIK = "rbUyusmazlik"
-    DUZENLEYICI = "rbDuzenleyici"
-    MAHKEME = "rbMahkeme"
-
-class KikSearchRequest(BaseModel):
-    """Model for KIK Decision search criteria."""
-    karar_tipi: KikKararTipi = Field(KikKararTipi.UYUSMAZLIK, description="Type")
-    karar_no: str = Field("", description="No")
-    karar_tarihi_baslangic: str = Field("", description="Start", pattern=r"^\d{2}\.\d{2}\.\d{4}$|^$")
-    karar_tarihi_bitis: str = Field("", description="End", pattern=r"^\d{2}\.\d{2}\.\d{4}$|^$")
-    resmi_gazete_sayisi: str = Field("", description="Gazette")
-    resmi_gazete_tarihi: str = Field("", description="Date", pattern=r"^\d{2}\.\d{2}\.\d{4}$|^$")
-    basvuru_konusu_ihale: str = Field("", description="Subject")
-    basvuru_sahibi: str = Field("", description="Applicant")
-    ihaleyi_yapan_idare: str = Field("", description="Entity")
-    yil: str = Field("", description="Year")
-    karar_metni: str = Field("", description="Text")
-    page: int = Field(1, ge=1, description="Page")
-
-class KikDecisionEntry(BaseModel):
-    """Represents a single decision entry from KIK search results."""
-    preview_event_target: str = Field(..., description="Event target")
-    karar_no_str: str = Field(..., alias="kararNo", description="Decision number")
-    karar_tipi: KikKararTipi = Field(..., description="Decision type")
-    
-    karar_tarihi_str: str = Field(..., alias="kararTarihi", description="Date")
-    idare_str: str = Field("", alias="idare", description="Entity")
-    basvuru_sahibi_str: str = Field("", alias="basvuruSahibi", description="Applicant")
-    ihale_konusu_str: str = Field("", alias="ihaleKonusu", description="Subject")
-
-    @computed_field
-    @property
-    def karar_id(self) -> str:
-        """
-        A Base64 encoded unique ID for the decision, combining decision type and number.
-        Format before encoding: "{karar_tipi.value}|{karar_no_str}"
-        """
-        combined_key = f"{self.karar_tipi.value}|{self.karar_no_str}"
-        return base64.b64encode(combined_key.encode('utf-8')).decode('utf-8')
-
-    model_config = ConfigDict(populate_by_name=True)
-
-class KikSearchResult(BaseModel):
-    """Model for KIK search results."""
-    decisions: List[KikDecisionEntry]
-    total_records: int = 0
-    current_page: int = 1
-
-class KikDocumentMarkdown(BaseModel):
-    """
-    KIK decision document, with Markdown content potentially paginated.
-    """
-    retrieved_with_karar_id: Optional[str] = Field(None, description="Request ID")
-    retrieved_karar_no: Optional[str] = Field(None, description="Decision number")
-    retrieved_karar_tipi: Optional[KikKararTipi] = Field(None, description="Decision type")
-    
-    karar_id_param_from_url: Optional[str] = Field(None, alias="kararIdParam", description="Internal ID")
-    markdown_chunk: Optional[str] = Field(None, description="Content")
-    source_url: Optional[str] = Field(None, description="Source URL")
-    error_message: Optional[str] = Field(None, description="Error")
-    current_page: int = Field(1, description="Page")
-    total_pages: int = Field(1, description="Total pages")
-    is_paginated: bool = Field(False, description="Paginated")
-    full_content_char_count: Optional[int] = Field(None, description="Char count")
-
-    model_config = ConfigDict(populate_by_name=True)

kik_mcp_module/models_v2.py

@@ -0,0 +1,147 @@
+# kik_mcp_module/models_v2.py
+from pydantic import BaseModel, Field, ConfigDict
+from typing import List, Optional
+from datetime import datetime
+from enum import Enum
+
+# New KIK v2 API Models
+
+class KikV2DecisionType(str, Enum):
+    """KIK v2 Decision Types with corresponding endpoints."""
+    UYUSMAZLIK = "uyusmazlik"        # Disputes - GetKurulKararlari
+    DUZENLEYICI = "duzenleyici"      # Regulatory - GetKurulKararlariDk  
+    MAHKEME = "mahkeme"              # Court - GetKurulKararlariMk
+
+class KikV2SearchRequest(BaseModel):
+    """Model for KIK v2 API search request."""
+    KararMetni: str = Field("", description="Decision text search query")
+    KararNo: str = Field("", description="Decision number (e.g., '2025/UH.II-1801')")
+    BasvuranAdi: str = Field("", description="Applicant name")
+    IdareAdi: str = Field("", description="Administration name")
+    BaslangicTarihi: str = Field("", description="Start date (YYYY-MM-DD)")
+    BitisTarihi: str = Field("", description="End date (YYYY-MM-DD)")
+    
+class KikV2KeyValuePair(BaseModel):
+    """Key-value pair for KIK v2 API request."""
+    key: str
+    value: str
+
+class KikV2QueryRequest(BaseModel):
+    """Nested query structure for KIK v2 API."""
+    keyValueOfstringanyType: List[KikV2KeyValuePair]
+
+class KikV2RequestData(BaseModel):
+    """Main request data structure for KIK v2 API."""
+    keyValuePairs: KikV2QueryRequest
+
+# Request Payloads for different decision types
+class KikV2SearchPayload(BaseModel):
+    """Complete payload for KIK v2 API search - Uyuşmazlık (Disputes)."""
+    sorgulaKurulKararlari: KikV2RequestData
+
+class KikV2SearchPayloadDk(BaseModel):
+    """Complete payload for KIK v2 API search - Düzenleyici (Regulatory)."""
+    sorgulaKurulKararlariDk: KikV2RequestData
+
+class KikV2SearchPayloadMk(BaseModel):
+    """Complete payload for KIK v2 API search - Mahkeme (Court)."""
+    sorgulaKurulKararlariMk: KikV2RequestData
+
+# Response Models
+
+class KikV2DecisionDetail(BaseModel):
+    """Individual decision detail from KIK v2 API response."""
+    resmiGazeteMukerrerSayi: str = Field("", description="Official Gazette duplicate number")
+    itiraz: str = Field("", description="Objection")
+    yayinlanmaTarihi: str = Field("", description="Publication date")
+    idareAdi: str = Field("", description="Administration name")
+    uzmanTCKN: str = Field("", description="Expert TCKN")
+    resmiGazeteTarihi: str = Field("", description="Official Gazette date")
+    basvuruKonusu: str = Field("", description="Application subject")
+    kararTurKod: str = Field("", description="Decision type code")
+    kararTurAciklama: str = Field("", description="Decision type description")
+    karar: str = Field("", description="Decision text")
+    kararNo: str = Field("", description="Decision number")
+    resmiGazeteSayisi: str = Field("", description="Official Gazette number")
+    inceleme: str = Field("", description="Review")
+    basvuruTarihi: str = Field("", description="Application date")
+    kararNitelikKod: str = Field("", description="Decision nature code")
+    resmiGazeteMukerrer: str = Field("", description="Official Gazette duplicate")
+    basvuruSayisi: str = Field("", description="Application number")
+    basvuran: str = Field("", description="Applicant")
+    kararNitelik: str = Field("", description="Decision nature")
+    uyusmazlikKararNo: str = Field("", description="Dispute decision number")
+    kurulNo: str = Field("", description="Board number")
+    gundemMaddesiSiraNo: str = Field("", description="Agenda item sequence")
+    kararTarihi: str = Field("", description="Decision date (ISO format)")
+    dosyaBirimKodu: str = Field("", description="File unit code")
+    gundemMaddesiId: str = Field("", description="Agenda item ID")
+
+class KikV2DecisionGroup(BaseModel):
+    """Group of decision details."""
+    KurulKararTutanakDetayi: List[KikV2DecisionDetail] = Field(alias="kurulKararTutanakDetayi")
+    
+    model_config = ConfigDict(populate_by_name=True)
+
+class KikV2SearchResultData(BaseModel):
+    """Search result data structure."""
+    hataKodu: str = Field("", description="Error code")
+    hataMesaji: str = Field("", description="Error message") 
+    KurulKararTutanakDetayListesi: List[KikV2DecisionGroup]
+    
+    model_config = ConfigDict(populate_by_name=True)
+
+class KikV2SearchResultWrapper(BaseModel):
+    """Wrapper for search result."""
+    SorgulaKurulKararlariResult: KikV2SearchResultData
+
+# Base Response Models
+class KikV2SearchResponse(BaseModel):
+    """Complete KIK v2 API search response for Uyuşmazlık (Disputes)."""
+    SorgulaKurulKararlariResponse: KikV2SearchResultWrapper
+
+# Düzenleyici Kararlar (Regulatory Decisions) Response Models
+class KikV2SearchResultWrapperDk(BaseModel):
+    """Wrapper for regulatory decisions search result."""
+    SorgulaKurulKararlariDkResult: KikV2SearchResultData
+
+class KikV2SearchResponseDk(BaseModel):
+    """Complete KIK v2 API search response for Düzenleyici (Regulatory) decisions."""
+    SorgulaKurulKararlariDkResponse: KikV2SearchResultWrapperDk
+
+# Mahkeme Kararlar (Court Decisions) Response Models  
+class KikV2SearchResultWrapperMk(BaseModel):
+    """Wrapper for court decisions search result."""
+    SorgulaKurulKararlariMkResult: KikV2SearchResultData
+
+class KikV2SearchResponseMk(BaseModel):
+    """Complete KIK v2 API search response for Mahkeme (Court) decisions."""
+    SorgulaKurulKararlariMkResponse: KikV2SearchResultWrapperMk
+
+# Simplified Models for MCP Tools
+
+class KikV2CompactDecision(BaseModel):
+    """Compact decision format for MCP tool responses."""
+    kararNo: str = Field("", description="Decision number")
+    kararTarihi: str = Field("", description="Decision date")
+    basvuran: str = Field("", description="Applicant")
+    idareAdi: str = Field("", description="Administration")
+    basvuruKonusu: str = Field("", description="Application subject")
+    gundemMaddesiId: str = Field("", description="Document ID for retrieval")
+    decision_type: str = Field("", description="Decision type (uyusmazlik/duzenleyici/mahkeme)")
+    
+class KikV2SearchResult(BaseModel):
+    """Compact search results for MCP tools."""
+    decisions: List[KikV2CompactDecision]
+    total_records: int = Field(0, description="Total number of decisions found")
+    page: int = Field(1, description="Current page number")
+    error_code: str = Field("", description="API error code")
+    error_message: str = Field("", description="API error message")
+
+class KikV2DocumentMarkdown(BaseModel):
+    """Document content in Markdown format."""
+    document_id: str = Field("", description="Document ID")
+    kararNo: str = Field("", description="Decision number")
+    markdown_content: str = Field("", description="Decision content in Markdown")
+    source_url: str = Field("", description="Source URL")
+    error_message: str = Field("", description="Error message if retrieval failed")

kvkk_mcp_module/client.py

@@ -1,5 +1,6 @@
 # kvkk_mcp_module/client.py
 
+import asyncio
 import httpx
 from bs4 import BeautifulSoup
 from typing import List, Optional, Dict, Any
@@ -291,7 +292,7 @@ class KvkkApiClient:
             # Convert HTML content to Markdown
             full_markdown_content = None
             if extracted_data["html_content"]:
-                full_markdown_content = self._convert_html_to_markdown(extracted_data["html_content"])
+                full_markdown_content = await asyncio.to_thread(self._convert_html_to_markdown, extracted_data["html_content"])
             
             if not full_markdown_content:
                 return KvkkDocumentMarkdown(

mcp_auth/__init__.py

@@ -1,28 +0,0 @@
-"""
-MCP Auth Toolkit - OAuth 2.1 + Authorization for Model Context Protocol Servers
-Integrated with Clerk Authentication
-"""
-
-from .middleware import (
-    AuthContext,
-    FastMCPAuthWrapper,
-    MCPAuthMiddleware,
-    auth_required,
-)
-from .oauth import OAuthConfig, OAuthProvider
-from .policy import PolicyEngine, ToolPolicy, create_default_policies
-from .storage import PersistentStorage
-
-__version__ = "0.1.0"
-__all__ = [
-    "OAuthProvider",
-    "OAuthConfig", 
-    "AuthContext",
-    "auth_required",
-    "create_default_policies",
-    "MCPAuthMiddleware",
-    "FastMCPAuthWrapper",
-    "PolicyEngine",
-    "ToolPolicy",
-    "PersistentStorage",
-]

mcp_auth/clerk_config.py

@@ -1,73 +0,0 @@
-"""
-Clerk OAuth configuration for MCP Auth Toolkit
-"""
-
-import os
-import logging
-from .oauth import OAuthConfig
-
-logger = logging.getLogger(__name__)
-
-
-def create_clerk_oauth_config() -> OAuthConfig:
-    """Create OAuth configuration for Clerk integration using SDK"""
-    
-    # Get Clerk configuration from environment
-    clerk_domain = os.getenv("CLERK_DOMAIN", "accounts.yargimcp.com")
-    clerk_publishable_key = os.getenv("CLERK_PUBLISHABLE_KEY")
-    clerk_secret_key = os.getenv("CLERK_SECRET_KEY")
-    
-    if not clerk_publishable_key or not clerk_secret_key:
-        raise ValueError("CLERK_PUBLISHABLE_KEY and CLERK_SECRET_KEY are required")
-    
-    # For Clerk with custom domains, we use our adapter endpoints
-    # This allows us to handle the custom domain flow properly
-    base_url = os.getenv("BASE_URL", "https://yargimcp.com")
-    
-    config = OAuthConfig(
-        client_id=clerk_publishable_key,
-        client_secret=clerk_secret_key,
-        # Use our adapter endpoints instead of Clerk's direct endpoints
-        authorization_endpoint=f"{base_url}/authorize",
-        token_endpoint=f"{base_url}/token",
-        # Keep Clerk's JWKS for token validation
-        jwks_uri=f"https://{clerk_domain}/.well-known/jwks.json",
-        issuer=base_url,  # We're the issuer for MCP tokens
-        scopes=["mcp:tools:read", "mcp:tools:write", "openid", "profile", "email"]
-    )
-    
-    logger.info(f"Created Clerk OAuth config with adapter endpoints")
-    logger.info(f"Clerk domain: {clerk_domain}")
-    logger.debug(f"Authorization endpoint: {config.authorization_endpoint}")
-    logger.debug(f"Token endpoint: {config.token_endpoint}")
-    
-    return config
-
-
-def get_jwt_secret() -> str:
-    """Get JWT secret for token signing"""
-    jwt_secret = os.getenv("JWT_SECRET_KEY")
-    
-    if not jwt_secret:
-        raise ValueError("JWT_SECRET_KEY environment variable is required")
-    
-    return jwt_secret
-
-
-def create_mcp_server_config():
-    """Create complete MCP server configuration for Clerk integration"""
-    
-    try:
-        oauth_config = create_clerk_oauth_config()
-        jwt_secret = get_jwt_secret()
-        
-        return {
-            "oauth_config": oauth_config,
-            "jwt_secret": jwt_secret,
-            "base_url": os.getenv("BASE_URL", "https://yargi-mcp.fly.dev"),
-            "auth_enabled": os.getenv("ENABLE_AUTH", "true").lower() == "true"
-        }
-        
-    except Exception as e:
-        logger.error(f"Failed to create MCP server config: {e}")
-        raise

mcp_auth/middleware.py

@@ -1,315 +0,0 @@
-"""
-MCP server middleware for OAuth authentication and authorization
-"""
-
-import functools
-import logging
-from collections.abc import Callable
-from dataclasses import dataclass
-from typing import Any, Optional
-
-logger = logging.getLogger(__name__)
-
-try:
-    from fastmcp import FastMCP
-    FASTMCP_AVAILABLE = True
-except ImportError:
-    FASTMCP_AVAILABLE = False
-    FastMCP = None
-    logger.warning("FastMCP not available, some features will be disabled")
-
-from .oauth import OAuthProvider
-from .policy import PolicyEngine
-
-
-@dataclass
-class AuthContext:
-    """Authentication context passed to MCP tools"""
-
-    user_id: str
-    scopes: list[str]
-    claims: dict[str, Any]
-    token: str
-
-
-class MCPAuthMiddleware:
-    """Authentication middleware for MCP servers"""
-
-    def __init__(self, oauth_provider: OAuthProvider, policy_engine: PolicyEngine):
-        self.oauth_provider = oauth_provider
-        self.policy_engine = policy_engine
-
-    def authenticate_request(self, authorization_header: str) -> AuthContext | None:
-        """Extract and validate auth token from request"""
-
-        if not authorization_header:
-            logger.debug("No authorization header provided")
-            return None
-
-        if not authorization_header.startswith("Bearer "):
-            logger.debug("Authorization header does not start with 'Bearer '")
-            return None
-
-        token = authorization_header[7:]  # Remove 'Bearer ' prefix
-
-        token_info = self.oauth_provider.introspect_token(token)
-
-        if not token_info.get("active"):
-            logger.warning("Token is not active")
-            return None
-
-        logger.debug(f"Authenticated user: {token_info.get('sub', 'unknown')}")
-
-        return AuthContext(
-            user_id=token_info.get("sub", "unknown"),
-            scopes=token_info.get("mcp_tool_scopes", []),
-            claims=token_info,
-            token=token,
-        )
-
-    def authorize_tool_call(
-        self, tool_name: str, auth_context: AuthContext
-    ) -> tuple[bool, str | None]:
-        """Check if user can call the specified tool"""
-
-        return self.policy_engine.authorize_tool_call(
-            tool_name=tool_name,
-            user_scopes=auth_context.scopes,
-            user_claims=auth_context.claims,
-        )
-
-
-def auth_required(
-    oauth_provider: OAuthProvider,
-    policy_engine: PolicyEngine,
-    tool_name: str | None = None,
-):
-    """
-    Decorator to require authentication for MCP tool functions
-
-    Usage:
-        @auth_required(oauth_provider, policy_engine, "search_yargitay")
-        def my_tool_function(context: AuthContext, ...):
-            pass
-    """
-
-    def decorator(func: Callable) -> Callable:
-        middleware = MCPAuthMiddleware(oauth_provider, policy_engine)
-
-        @functools.wraps(func)
-        async def wrapper(*args, **kwargs):
-            # Extract authorization header from kwargs
-            auth_header = kwargs.pop("authorization", None)
-            
-            # Also check in args if it's a Request object
-            if not auth_header and args:
-                for arg in args:
-                    if hasattr(arg, 'headers'):
-                        auth_header = arg.headers.get("Authorization")
-                        break
-
-            if not auth_header:
-                logger.warning(f"No authorization header for tool '{tool_name or func.__name__}'")
-                raise PermissionError("Authorization header required")
-
-            auth_context = middleware.authenticate_request(auth_header)
-
-            if not auth_context:
-                logger.warning(f"Authentication failed for tool '{tool_name or func.__name__}'")
-                raise PermissionError("Invalid or expired token")
-
-            actual_tool_name = tool_name or func.__name__
-
-            authorized, reason = middleware.authorize_tool_call(
-                actual_tool_name, auth_context
-            )
-
-            if not authorized:
-                logger.warning(f"Authorization failed for tool '{actual_tool_name}': {reason}")
-                raise PermissionError(f"Access denied: {reason}")
-
-            # Add auth context to function call
-            return await func(auth_context, *args, **kwargs)
-
-        return wrapper
-
-    return decorator
-
-
-class FastMCPAuthWrapper:
-    """Wrapper for FastMCP servers to add authentication"""
-
-    def __init__(
-        self,
-        mcp_server: "FastMCP",
-        oauth_provider: OAuthProvider,
-        policy_engine: PolicyEngine,
-    ):
-        if not FASTMCP_AVAILABLE:
-            raise ImportError("FastMCP is required for FastMCPAuthWrapper")
-            
-        self.mcp_server = mcp_server
-        self.middleware = MCPAuthMiddleware(oauth_provider, policy_engine)
-        self.oauth_provider = oauth_provider
-        logger.info("Initializing FastMCP authentication wrapper")
-        self._wrap_tools()
-
-    def _wrap_tools(self):
-        """Wrap all existing tools with auth middleware"""
-
-        # Try different FastMCP tool storage locations
-        tool_registry = None
-        
-        if hasattr(self.mcp_server, '_tools'):
-            tool_registry = self.mcp_server._tools
-        elif hasattr(self.mcp_server, 'tools'):
-            tool_registry = self.mcp_server.tools
-        elif hasattr(self.mcp_server, '_tool_registry'):
-            tool_registry = self.mcp_server._tool_registry
-        elif hasattr(self.mcp_server, '_handlers') and hasattr(self.mcp_server._handlers, 'tools'):
-            tool_registry = self.mcp_server._handlers.tools
-        
-        if not tool_registry:
-            logger.warning("FastMCP server tool registry not found, tools will not be automatically wrapped")
-            logger.debug(f"Available server attributes: {dir(self.mcp_server)}")
-            return
-
-        logger.debug(f"Found tool registry with {len(tool_registry)} tools")
-        original_tools = dict(tool_registry)
-        wrapped_count = 0
-
-        for tool_name, tool_func in original_tools.items():
-            try:
-                wrapped_func = self._create_auth_wrapper(tool_name, tool_func)
-                tool_registry[tool_name] = wrapped_func
-                wrapped_count += 1
-                logger.debug(f"Wrapped tool: {tool_name}")
-            except Exception as e:
-                logger.error(f"Failed to wrap tool {tool_name}: {e}")
-
-        logger.info(f"Successfully wrapped {wrapped_count} tools with authentication")
-
-    def _create_auth_wrapper(self, tool_name: str, original_func: Callable) -> Callable:
-        """Create auth wrapper for a specific tool"""
-
-        @functools.wraps(original_func)
-        async def auth_wrapper(*args, **kwargs):
-            # Extract authorization from various sources
-            auth_header = None
-            
-            # Check kwargs first
-            auth_header = kwargs.pop("authorization", None)
-            
-            # Check if first argument is a Request object
-            if not auth_header and args:
-                first_arg = args[0]
-                if hasattr(first_arg, 'headers'):
-                    auth_header = first_arg.headers.get("Authorization")
-
-            if not auth_header:
-                logger.warning(f"No authorization header for tool '{tool_name}'")
-                raise PermissionError("Authorization required")
-
-            auth_context = self.middleware.authenticate_request(auth_header)
-
-            if not auth_context:
-                logger.warning(f"Authentication failed for tool '{tool_name}'")
-                raise PermissionError("Invalid token")
-
-            authorized, reason = self.middleware.authorize_tool_call(
-                tool_name, auth_context
-            )
-
-            if not authorized:
-                logger.warning(f"Authorization failed for tool '{tool_name}': {reason}")
-                raise PermissionError(f"Access denied: {reason}")
-
-            # Add auth context to kwargs
-            kwargs["auth_context"] = auth_context
-            logger.debug(f"Calling tool '{tool_name}' for user {auth_context.user_id}")
-            
-            return await original_func(*args, **kwargs)
-
-        return auth_wrapper
-
-    def add_oauth_endpoints(self):
-        """Add OAuth endpoints to the MCP server"""
-
-        @self.mcp_server.tool(
-            description="Initiate OAuth 2.1 authorization flow with PKCE",
-            annotations={"readOnlyHint": True, "idempotentHint": False}
-        )
-        async def oauth_authorize(redirect_uri: str, scopes: Optional[str] = None):
-            """OAuth authorization endpoint"""
-            scope_list = scopes.split(" ") if scopes else None
-            auth_url, pkce = self.oauth_provider.generate_authorization_url(
-                redirect_uri=redirect_uri, scopes=scope_list
-            )
-            logger.info(f"Generated authorization URL for redirect_uri: {redirect_uri}")
-            return {
-                "authorization_url": auth_url,
-                "code_verifier": pkce.verifier,  # For PKCE flow
-                "code_challenge": pkce.challenge,
-                "instructions": "Use the authorization_url to complete OAuth flow, then exchange the returned code using oauth_token tool"
-            }
-
-        @self.mcp_server.tool(
-            description="Exchange OAuth authorization code for access token",
-            annotations={"readOnlyHint": False, "idempotentHint": False}
-        )
-        async def oauth_token(
-            code: str,
-            state: str,
-            redirect_uri: str
-        ):
-            """OAuth token exchange endpoint"""
-            try:
-                result = await self.oauth_provider.exchange_code_for_token(
-                    code=code, state=state, redirect_uri=redirect_uri
-                )
-                logger.info("Successfully exchanged authorization code for token")
-                return result
-            except Exception as e:
-                logger.error(f"Token exchange failed: {e}")
-                raise
-
-        @self.mcp_server.tool(
-            description="Validate and introspect OAuth access token",
-            annotations={"readOnlyHint": True, "idempotentHint": True}
-        )
-        async def oauth_introspect(token: str):
-            """Token introspection endpoint"""
-            result = self.oauth_provider.introspect_token(token)
-            logger.debug(f"Token introspection: active={result.get('active', False)}")
-            return result
-
-        @self.mcp_server.tool(
-            description="Revoke OAuth access token",
-            annotations={"readOnlyHint": False, "idempotentHint": False}
-        )
-        async def oauth_revoke(token: str):
-            """Token revocation endpoint"""
-            success = self.oauth_provider.revoke_token(token)
-            logger.info(f"Token revocation: success={success}")
-            return {"revoked": success}
-
-        @self.mcp_server.tool(
-            description="Get list of tools available to authenticated user",
-            annotations={"readOnlyHint": True, "idempotentHint": True}
-        )
-        async def oauth_user_tools(authorization: str):
-            """Get user's allowed tools based on scopes"""
-            auth_context = self.middleware.authenticate_request(authorization)
-            if not auth_context:
-                raise PermissionError("Invalid token")
-            
-            allowed_patterns = self.middleware.policy_engine.get_allowed_tools(auth_context.scopes)
-            
-            return {
-                "user_id": auth_context.user_id,
-                "scopes": auth_context.scopes,
-                "allowed_tool_patterns": allowed_patterns,
-                "message": "Use these patterns to determine which tools you can access"
-            }
-
-        logger.info("Added OAuth endpoints: oauth_authorize, oauth_token, oauth_introspect, oauth_revoke, oauth_user_tools")

mcp_auth/oauth.py

@@ -1,304 +0,0 @@
-"""
-OAuth 2.1 + PKCE implementation for MCP servers with Clerk integration
-"""
-
-import base64
-import hashlib
-import secrets
-import time
-import logging
-from dataclasses import dataclass
-from datetime import datetime, timedelta
-from typing import Any, Optional
-from urllib.parse import urlencode
-
-import httpx
-import jwt
-from jwt.exceptions import PyJWTError, InvalidTokenError
-
-from .storage import PersistentStorage
-
-# Try to import Clerk SDK
-try:
-    from clerk_backend_api import Clerk
-    CLERK_AVAILABLE = True
-except ImportError:
-    CLERK_AVAILABLE = False
-    Clerk = None
-
-logger = logging.getLogger(__name__)
-
-
-@dataclass
-class OAuthConfig:
-    """OAuth provider configuration for Clerk"""
-
-    client_id: str
-    client_secret: str
-    authorization_endpoint: str
-    token_endpoint: str
-    jwks_uri: str | None = None
-    issuer: str = "mcp-auth"
-    scopes: list[str] = None
-
-    def __post_init__(self):
-        if self.scopes is None:
-            self.scopes = ["mcp:tools:read", "mcp:tools:write"]
-
-
-class PKCEChallenge:
-    """PKCE challenge/verifier pair for OAuth 2.1"""
-
-    def __init__(self):
-        self.verifier = (
-            base64.urlsafe_b64encode(secrets.token_bytes(32))
-            .decode("utf-8")
-            .rstrip("=")
-        )
-
-        challenge_bytes = hashlib.sha256(self.verifier.encode("utf-8")).digest()
-        self.challenge = (
-            base64.urlsafe_b64encode(challenge_bytes).decode("utf-8").rstrip("=")
-        )
-
-
-class OAuthProvider:
-    """OAuth 2.1 provider with PKCE support and Clerk integration"""
-
-    def __init__(self, config: OAuthConfig, jwt_secret: str):
-        self.config = config
-        self.jwt_secret = jwt_secret
-        # Use persistent storage instead of memory
-        self.storage = PersistentStorage()
-        
-        # Initialize Clerk SDK if available
-        self.clerk = None
-        if CLERK_AVAILABLE and config.client_secret:
-            try:
-                self.clerk = Clerk(bearer_auth=config.client_secret)
-                logger.info("Clerk SDK initialized successfully")
-            except Exception as e:
-                logger.warning(f"Failed to initialize Clerk SDK: {e}")
-                
-        logger.info("OAuth provider initialized with persistent storage")
-
-    def generate_authorization_url(
-        self,
-        redirect_uri: str,
-        state: str | None = None,
-        scopes: list[str] | None = None,
-    ) -> tuple[str, PKCEChallenge]:
-        """Generate OAuth authorization URL with PKCE for Clerk"""
-
-        pkce = PKCEChallenge()
-        session_id = secrets.token_urlsafe(32)
-
-        if state is None:
-            state = secrets.token_urlsafe(16)
-
-        if scopes is None:
-            scopes = self.config.scopes
-
-        # Store session data with expiration
-        session_data = {
-            "pkce_verifier": pkce.verifier,
-            "state": state,
-            "redirect_uri": redirect_uri,
-            "scopes": scopes,
-            "created_at": time.time(),
-            "expires_at": (datetime.utcnow() + timedelta(minutes=10)).timestamp(),
-        }
-        self.storage.set_session(session_id, session_data)
-
-        # Build Clerk OAuth URL
-        # Check if this is a custom domain (sign-in endpoint)
-        if self.config.authorization_endpoint.endswith('/sign-in'):
-            # For custom domains, Clerk expects redirect_url parameter
-            params = {
-                "redirect_url": redirect_uri,
-                "state": f"{state}:{session_id}",
-            }
-            auth_url = f"{self.config.authorization_endpoint}?{urlencode(params)}"
-        else:
-            # Standard OAuth flow with PKCE
-            params = {
-                "response_type": "code",
-                "client_id": self.config.client_id,
-                "redirect_uri": redirect_uri,
-                "scope": " ".join(scopes),
-                "state": f"{state}:{session_id}",  # Combine state with session ID
-                "code_challenge": pkce.challenge,
-                "code_challenge_method": "S256",
-            }
-            auth_url = f"{self.config.authorization_endpoint}?{urlencode(params)}"
-        
-        logger.info(f"Generated OAuth URL with session {session_id[:8]}...")
-        logger.debug(f"Auth URL: {auth_url}")
-        return auth_url, pkce
-
-    async def exchange_code_for_token(
-        self, code: str, state: str, redirect_uri: str
-    ) -> dict[str, Any]:
-        """Exchange authorization code for access token with Clerk"""
-
-        try:
-            original_state, session_id = state.split(":", 1)
-        except ValueError as e:
-            logger.error(f"Invalid state format: {state}")
-            raise ValueError("Invalid state format") from e
-
-        session = self.storage.get_session(session_id)
-        if not session:
-            logger.error(f"Session {session_id} not found")
-            raise ValueError("Invalid session")
-        
-        # Check session expiration
-        if datetime.utcnow().timestamp() > session.get("expires_at", 0):
-            self.storage.delete_session(session_id)
-            logger.error(f"Session {session_id} expired")
-            raise ValueError("Session expired")
-
-        if session["state"] != original_state:
-            logger.error(f"State mismatch: expected {session['state']}, got {original_state}")
-            raise ValueError("State mismatch")
-
-        if session["redirect_uri"] != redirect_uri:
-            logger.error(f"Redirect URI mismatch: expected {session['redirect_uri']}, got {redirect_uri}")
-            raise ValueError("Redirect URI mismatch")
-
-        # Prepare token exchange request for Clerk
-        token_data = {
-            "grant_type": "authorization_code",
-            "client_id": self.config.client_id,
-            "client_secret": self.config.client_secret,
-            "code": code,
-            "redirect_uri": redirect_uri,
-            "code_verifier": session["pkce_verifier"],
-        }
-
-        logger.info(f"Exchanging code with Clerk for session {session_id[:8]}...")
-
-        async with httpx.AsyncClient() as client:
-            response = await client.post(
-                self.config.token_endpoint,
-                data=token_data,
-                headers={"Content-Type": "application/x-www-form-urlencoded"},
-                timeout=30.0,
-            )
-
-        if response.status_code != 200:
-            logger.error(f"Clerk token exchange failed: {response.status_code} - {response.text}")
-            raise ValueError(f"Token exchange failed: {response.text}")
-
-        token_response = response.json()
-        logger.info("Successfully exchanged code for Clerk token")
-
-        # Create MCP-scoped JWT token
-        access_token = self._create_mcp_token(
-            session["scopes"], token_response.get("access_token"), session_id
-        )
-
-        # Store token for introspection
-        token_id = secrets.token_urlsafe(16)
-        token_data = {
-            "access_token": access_token,
-            "scopes": session["scopes"],
-            "created_at": time.time(),
-            "expires_at": (datetime.utcnow() + timedelta(hours=1)).timestamp(),
-            "session_id": session_id,
-            "clerk_token": token_response.get("access_token"),
-        }
-        self.storage.set_token(token_id, token_data)
-
-        # Clean up session
-        self.storage.delete_session(session_id)
-
-        return {
-            "access_token": access_token,
-            "token_type": "bearer",
-            "expires_in": 3600,
-            "scope": " ".join(session["scopes"]),
-        }
-
-    def validate_pkce(self, code_verifier: str, code_challenge: str) -> bool:
-        """Validate PKCE code challenge (RFC 7636)"""
-        # S256 method
-        verifier_hash = hashlib.sha256(code_verifier.encode()).digest()
-        expected_challenge = base64.urlsafe_b64encode(verifier_hash).decode().rstrip('=')
-        return expected_challenge == code_challenge
-
-    def _create_mcp_token(
-        self, scopes: list[str], upstream_token: str, session_id: str
-    ) -> str:
-        """Create MCP-scoped JWT token with Clerk token embedded"""
-
-        now = int(time.time())
-        payload = {
-            "iss": self.config.issuer,
-            "sub": session_id,
-            "aud": "mcp-server",
-            "iat": now,
-            "exp": now + 3600,  # 1 hour expiration
-            "mcp_tool_scopes": scopes,
-            "upstream_token": upstream_token,
-            "clerk_integration": True,
-        }
-
-        return jwt.encode(payload, self.jwt_secret, algorithm="HS256")
-
-    def introspect_token(self, token: str) -> dict[str, Any]:
-        """Introspect and validate MCP token"""
-
-        try:
-            payload = jwt.decode(token, self.jwt_secret, algorithms=["HS256"])
-
-            # Check if token is expired
-            if payload.get("exp", 0) < time.time():
-                return {"active": False, "error": "token_expired"}
-
-            return {
-                "active": True,
-                "sub": payload.get("sub"),
-                "aud": payload.get("aud"),
-                "iss": payload.get("iss"),
-                "exp": payload.get("exp"),
-                "iat": payload.get("iat"),
-                "mcp_tool_scopes": payload.get("mcp_tool_scopes", []),
-                "upstream_token": payload.get("upstream_token"),
-                "clerk_integration": payload.get("clerk_integration", False),
-            }
-
-        except PyJWTError as e:
-            logger.warning(f"Token validation failed: {e}")
-            return {"active": False, "error": "invalid_token"}
-
-    def revoke_token(self, token: str) -> bool:
-        """Revoke a token"""
-
-        try:
-            payload = jwt.decode(token, self.jwt_secret, algorithms=["HS256"])
-            session_id = payload.get("sub")
-
-            # Remove all tokens associated with this session
-            all_tokens = self.storage.get_tokens()
-            tokens_to_remove = [
-                token_id
-                for token_id, token_data in all_tokens.items()
-                if token_data.get("session_id") == session_id
-            ]
-
-            for token_id in tokens_to_remove:
-                self.storage.delete_token(token_id)
-
-            logger.info(f"Revoked {len(tokens_to_remove)} tokens for session {session_id}")
-            return True
-
-        except InvalidTokenError as e:
-            logger.warning(f"Token revocation failed: {e}")
-            return False
-
-    def cleanup_expired_sessions(self):
-        """Clean up expired sessions and tokens"""
-        # This is now handled automatically by persistent storage
-        self.storage.cleanup_expired_sessions()
-        logger.debug("Cleanup completed via persistent storage")

mcp_auth/policy.py

@@ -1,201 +0,0 @@
-"""
-Authorization policy engine for MCP tools
-"""
-
-import re
-import logging
-from dataclasses import dataclass
-from enum import Enum
-from typing import Any
-
-logger = logging.getLogger(__name__)
-
-
-class PolicyAction(Enum):
-    ALLOW = "allow"
-    DENY = "deny"
-
-
-@dataclass
-class ToolPolicy:
-    """Policy rule for MCP tool access"""
-
-    tool_pattern: str  # regex pattern for tool names
-    required_scopes: list[str]
-    action: PolicyAction = PolicyAction.ALLOW
-    conditions: dict[str, Any] | None = None
-
-    def matches_tool(self, tool_name: str) -> bool:
-        """Check if the policy applies to given tool"""
-        return bool(re.match(self.tool_pattern, tool_name))
-
-    def evaluate_scopes(self, user_scopes: list[str]) -> bool:
-        """Check if user has required scopes"""
-        return all(scope in user_scopes for scope in self.required_scopes)
-
-
-class PolicyEngine:
-    """Authorization policy engine for Turkish legal database tools"""
-
-    def __init__(self):
-        self.policies: list[ToolPolicy] = []
-        self.default_action = PolicyAction.DENY
-
-    def add_policy(self, policy: ToolPolicy):
-        """Add a policy rule"""
-        self.policies.append(policy)
-        logger.debug(f"Added policy: {policy.tool_pattern} -> {policy.required_scopes}")
-
-    def add_tool_scope_policy(
-        self,
-        tool_pattern: str,
-        required_scopes: str | list[str],
-        action: PolicyAction = PolicyAction.ALLOW,
-    ):
-        """Convenience method to add tool-scope policy"""
-        if isinstance(required_scopes, str):
-            required_scopes = [required_scopes]
-
-        policy = ToolPolicy(
-            tool_pattern=tool_pattern, required_scopes=required_scopes, action=action
-        )
-        self.add_policy(policy)
-
-    def authorize_tool_call(
-        self,
-        tool_name: str,
-        user_scopes: list[str],
-        user_claims: dict[str, Any] | None = None,
-    ) -> tuple[bool, str | None]:
-        """
-        Authorize a tool call
-
-        Returns:
-            (authorized: bool, reason: Optional[str])
-        """
-
-        logger.debug(f"Authorizing tool '{tool_name}' for user with scopes: {user_scopes}")
-
-        matching_policies = [
-            policy for policy in self.policies if policy.matches_tool(tool_name)
-        ]
-
-        if not matching_policies:
-            if self.default_action == PolicyAction.ALLOW:
-                logger.debug(f"No policies found for '{tool_name}', allowing by default")
-                return True, None
-            else:
-                logger.warning(f"No policies found for '{tool_name}', denying by default")
-                return False, f"No policy found for tool '{tool_name}', default deny"
-
-        # Check for explicit deny policies first
-        for policy in matching_policies:
-            if policy.action == PolicyAction.DENY:
-                if policy.evaluate_scopes(user_scopes):
-                    logger.warning(f"Explicit deny policy matched for '{tool_name}'")
-                    return False, f"Explicit deny policy for tool '{tool_name}'"
-
-        # Check allow policies
-        allow_policies = [
-            p for p in matching_policies if p.action == PolicyAction.ALLOW
-        ]
-
-        if not allow_policies:
-            logger.warning(f"No allow policies found for '{tool_name}'")
-            return False, f"No allow policies found for tool '{tool_name}'"
-
-        for policy in allow_policies:
-            if policy.evaluate_scopes(user_scopes):
-                if self._evaluate_conditions(policy.conditions, user_claims):
-                    logger.debug(f"Authorization granted for '{tool_name}'")
-                    return True, None
-
-        logger.warning(f"Insufficient scopes for '{tool_name}'. Required: {[p.required_scopes for p in allow_policies]}, User has: {user_scopes}")
-        return False, f"Insufficient scopes for tool '{tool_name}'"
-
-    def _evaluate_conditions(
-        self,
-        conditions: dict[str, Any] | None,
-        user_claims: dict[str, Any] | None,
-    ) -> bool:
-        """Evaluate additional policy conditions"""
-
-        if not conditions:
-            return True
-
-        if not user_claims:
-            logger.debug("No user claims provided, conditions evaluation failed")
-            return False
-
-        for key, expected_value in conditions.items():
-            user_value = user_claims.get(key)
-
-            if isinstance(expected_value, list):
-                if user_value not in expected_value:
-                    logger.debug(f"Condition failed: {key} = {user_value} not in {expected_value}")
-                    return False
-            elif user_value != expected_value:
-                logger.debug(f"Condition failed: {key} = {user_value} != {expected_value}")
-                return False
-
-        return True
-
-    def get_allowed_tools(self, user_scopes: list[str]) -> list[str]:
-        """Get list of tool patterns user is allowed to call"""
-
-        allowed_tools = []
-
-        for policy in self.policies:
-            if policy.action == PolicyAction.ALLOW and policy.evaluate_scopes(
-                user_scopes
-            ):
-                allowed_tools.append(policy.tool_pattern)
-
-        return allowed_tools
-
-
-def create_turkish_legal_policies() -> PolicyEngine:
-    """Create policy set for Turkish legal database MCP server"""
-
-    engine = PolicyEngine()
-
-    # Administrative tools (full access)
-    engine.add_tool_scope_policy(".*", ["mcp:tools:admin"])
-
-    # Search tools - require read access
-    engine.add_tool_scope_policy("search.*", ["mcp:tools:read"])
-    
-    # Fetch/get document tools - require read access  
-    engine.add_tool_scope_policy("get_.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("fetch.*", ["mcp:tools:read"])
-
-    # Specific Turkish legal database tools
-    engine.add_tool_scope_policy("search_yargitay.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_danistay.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_anayasa.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_rekabet.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_kik.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_emsal.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_uyusmazlik.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_sayistay.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_.*_bedesten", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_yerel_hukuk.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_istinaf_hukuk.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("search_kyb.*", ["mcp:tools:read"])
-
-    # Document retrieval tools
-    engine.add_tool_scope_policy("get_.*_document.*", ["mcp:tools:read"])
-    engine.add_tool_scope_policy("get_.*_markdown", ["mcp:tools:read"])
-
-    # Write operations (if any future tools need them)
-    engine.add_tool_scope_policy("create_.*", ["mcp:tools:write"])
-    engine.add_tool_scope_policy("update_.*", ["mcp:tools:write"]) 
-    engine.add_tool_scope_policy("delete_.*", ["mcp:tools:write"])
-
-    logger.info("Created Turkish legal database policy engine")
-    return engine
-
-
-def create_default_policies() -> PolicyEngine:
-    """Create a default policy set for MCP servers (backwards compatibility)"""
-    return create_turkish_legal_policies()

mcp_auth/storage.py

@@ -1,112 +0,0 @@
-"""
-Persistent storage for OAuth sessions and tokens
-"""
-
-import json
-import os
-import tempfile
-import logging
-from datetime import datetime
-from typing import Dict, Any, Optional
-
-logger = logging.getLogger(__name__)
-
-
-class PersistentStorage:
-    """File-based persistent storage for OAuth data"""
-    
-    def __init__(self, storage_dir: str = None):
-        if storage_dir is None:
-            # Use system temp directory or environment variable
-            storage_dir = os.environ.get('TEMP', tempfile.gettempdir())
-        
-        self.storage_dir = os.path.join(storage_dir, 'mcp_oauth_storage')
-        os.makedirs(self.storage_dir, exist_ok=True)
-        
-        self.sessions_file = os.path.join(self.storage_dir, 'oauth_sessions.json')
-        self.tokens_file = os.path.join(self.storage_dir, 'oauth_tokens.json')
-        
-        logger.info(f"Persistent OAuth storage initialized at: {self.storage_dir}")
-    
-    def _load_json(self, filepath: str) -> Dict:
-        """Load JSON data from file"""
-        try:
-            if os.path.exists(filepath):
-                with open(filepath, 'r', encoding='utf-8') as f:
-                    return json.load(f)
-        except Exception as e:
-            logger.error(f"Error loading {filepath}: {e}")
-        return {}
-    
-    def _save_json(self, filepath: str, data: Dict):
-        """Save JSON data to file"""
-        try:
-            with open(filepath, 'w', encoding='utf-8') as f:
-                json.dump(data, f, indent=2, default=str)
-        except Exception as e:
-            logger.error(f"Error saving {filepath}: {e}")
-    
-    def get_sessions(self) -> Dict[str, Dict[str, Any]]:
-        """Get all OAuth sessions"""
-        data = self._load_json(self.sessions_file)
-        # Clean expired sessions
-        now = datetime.utcnow().timestamp()
-        valid_sessions = {k: v for k, v in data.items() 
-                         if v.get('expires_at', 0) > now}
-        if len(valid_sessions) != len(data):
-            self._save_json(self.sessions_file, valid_sessions)
-        return valid_sessions
-    
-    def set_session(self, session_id: str, data: Dict[str, Any]):
-        """Set OAuth session data"""
-        sessions = self.get_sessions()
-        sessions[session_id] = data
-        self._save_json(self.sessions_file, sessions)
-    
-    def get_session(self, session_id: str) -> Optional[Dict[str, Any]]:
-        """Get specific OAuth session data"""
-        sessions = self.get_sessions()
-        return sessions.get(session_id)
-    
-    def delete_session(self, session_id: str):
-        """Delete OAuth session"""
-        sessions = self.get_sessions()
-        if session_id in sessions:
-            del sessions[session_id]
-            self._save_json(self.sessions_file, sessions)
-    
-    def get_tokens(self) -> Dict[str, Dict[str, Any]]:
-        """Get all OAuth tokens"""
-        data = self._load_json(self.tokens_file)
-        # Clean expired tokens
-        now = datetime.utcnow().timestamp()
-        valid_tokens = {k: v for k, v in data.items() 
-                       if v.get('expires_at', 0) > now}
-        if len(valid_tokens) != len(data):
-            self._save_json(self.tokens_file, valid_tokens)
-        return valid_tokens
-    
-    def set_token(self, token_id: str, token_data: Dict[str, Any]):
-        """Set OAuth token data"""
-        tokens = self.get_tokens()
-        tokens[token_id] = token_data
-        self._save_json(self.tokens_file, tokens)
-    
-    def get_token(self, token_id: str) -> Optional[Dict[str, Any]]:
-        """Get specific OAuth token data"""
-        tokens = self.get_tokens()
-        return tokens.get(token_id)
-    
-    def delete_token(self, token_id: str):
-        """Delete OAuth token"""
-        tokens = self.get_tokens()
-        if token_id in tokens:
-            del tokens[token_id]
-            self._save_json(self.tokens_file, tokens)
-    
-    def cleanup_expired_sessions(self):
-        """Clean up expired sessions and tokens"""
-        # This is handled automatically in get_sessions() and get_tokens()
-        sessions = self.get_sessions()
-        tokens = self.get_tokens()
-        logger.debug(f"Cleanup: {len(sessions)} active sessions, {len(tokens)} active tokens")

mcp_auth_factory.py

@@ -1,193 +0,0 @@
-"""
-Factory for creating FastMCP app with MCP Auth Toolkit integration
-"""
-
-import logging
-import os
-from typing import Optional
-
-logger = logging.getLogger(__name__)
-
-try:
-    from fastmcp import FastMCP
-    FASTMCP_AVAILABLE = True
-except ImportError:
-    FASTMCP_AVAILABLE = False
-    FastMCP = None
-
-from mcp_auth import (
-    OAuthProvider, 
-    PolicyEngine, 
-    FastMCPAuthWrapper,
-    create_default_policies
-)
-from mcp_auth.clerk_config import create_mcp_server_config
-
-
-def create_auth_enabled_app(app_name: str = "Yargı MCP Server") -> FastMCP:
-    """Create FastMCP app with authentication enabled"""
-    
-    if not FASTMCP_AVAILABLE:
-        raise ImportError("FastMCP is required for authenticated MCP server")
-    
-    logger.info("Creating FastMCP app with MCP Auth Toolkit integration")
-    
-    # Create base FastMCP app
-    app = FastMCP(app_name)
-    
-    # Check if authentication is enabled
-    auth_enabled = os.getenv("ENABLE_AUTH", "true").lower() == "true"
-    
-    if not auth_enabled:
-        logger.info("Authentication disabled, returning basic FastMCP app")
-        return app
-    
-    try:
-        # Get configuration
-        logger.info("Getting MCP server configuration...")
-        config = create_mcp_server_config()
-        logger.info("Configuration loaded successfully")
-        
-        # Create OAuth provider with Clerk config
-        logger.info("Creating OAuth provider...")
-        oauth_provider = OAuthProvider(
-            config=config["oauth_config"],
-            jwt_secret=config["jwt_secret"]
-        )
-        logger.info("OAuth provider created successfully")
-        
-        # Create policy engine for Turkish legal database
-        policy_engine = create_default_policies()
-        
-        # Store auth components for later wrapping (after tools are defined)
-        app._oauth_provider = oauth_provider
-        app._policy_engine = policy_engine
-        app._auth_config = config
-        
-        # Add OAuth endpoints immediately
-        @app.tool(
-            description="Initiate OAuth 2.1 authorization flow with PKCE",
-            annotations={"readOnlyHint": True, "idempotentHint": False}
-        )
-        async def oauth_authorize(redirect_uri: str, scopes: str = None):
-            """OAuth authorization endpoint"""
-            scope_list = scopes.split(" ") if scopes else ["mcp:tools:read", "mcp:tools:write"]
-            auth_url, pkce = oauth_provider.generate_authorization_url(
-                redirect_uri=redirect_uri, scopes=scope_list
-            )
-            logger.info(f"Generated authorization URL for redirect_uri: {redirect_uri}")
-            return {
-                "authorization_url": auth_url,
-                "code_verifier": pkce.verifier,
-                "code_challenge": pkce.challenge,
-                "instructions": "Use the authorization_url to complete OAuth flow, then exchange the returned code using oauth_token tool"
-            }
-
-        @app.tool(
-            description="Exchange OAuth authorization code for access token",
-            annotations={"readOnlyHint": False, "idempotentHint": False}
-        )
-        async def oauth_token(code: str, state: str, redirect_uri: str):
-            """OAuth token exchange endpoint"""
-            try:
-                result = await oauth_provider.exchange_code_for_token(
-                    code=code, state=state, redirect_uri=redirect_uri
-                )
-                logger.info("Successfully exchanged authorization code for token")
-                return result
-            except Exception as e:
-                logger.error(f"Token exchange failed: {e}")
-                raise
-
-        @app.tool(
-            description="Validate and introspect OAuth access token",
-            annotations={"readOnlyHint": True, "idempotentHint": True}
-        )
-        async def oauth_introspect(token: str):
-            """Token introspection endpoint"""
-            result = oauth_provider.introspect_token(token)
-            logger.debug(f"Token introspection: active={result.get('active', False)}")
-            return result
-
-        @app.tool(
-            description="Revoke OAuth access token",
-            annotations={"readOnlyHint": False, "idempotentHint": False}
-        )
-        async def oauth_revoke(token: str):
-            """Token revocation endpoint"""
-            success = oauth_provider.revoke_token(token)
-            logger.info(f"Token revocation: success={success}")
-            return {"revoked": success}
-        
-        logger.info("Successfully created authenticated FastMCP app")
-        
-    except Exception as e:
-        logger.error(f"Failed to create authenticated app: {e}")
-        logger.info("Falling back to non-authenticated FastMCP app")
-        # Return basic app if auth setup fails
-        return app
-    
-    return app
-
-
-def create_app() -> FastMCP:
-    """Create FastMCP app (backwards compatible with mcp_factory.py)"""
-    return create_auth_enabled_app()
-
-
-def get_auth_wrapper(app: FastMCP) -> Optional[FastMCPAuthWrapper]:
-    """Get auth wrapper from app if available"""
-    return getattr(app, '_auth_wrapper', None)
-
-
-def get_oauth_provider(app: FastMCP) -> Optional[OAuthProvider]:
-    """Get OAuth provider from app if available"""
-    return getattr(app, '_oauth_provider', None)
-
-
-def get_policy_engine(app: FastMCP) -> Optional[PolicyEngine]:
-    """Get policy engine from app if available"""
-    return getattr(app, '_policy_engine', None)
-
-
-def is_auth_enabled(app: FastMCP) -> bool:
-    """Check if authentication is enabled for the app"""
-    return hasattr(app, '_oauth_provider') or hasattr(app, '_auth_wrapper')
-
-
-def enable_tool_authentication(app: FastMCP):
-    """Enable authentication on all existing tools (call after tools are defined)"""
-    if not is_auth_enabled(app):
-        logger.debug("Authentication not enabled, skipping tool authentication")
-        return
-    
-    oauth_provider = get_oauth_provider(app)
-    policy_engine = get_policy_engine(app)
-    
-    if not oauth_provider or not policy_engine:
-        logger.warning("OAuth provider or policy engine not available")
-        return
-    
-    try:
-        # Create auth wrapper and wrap tools
-        auth_wrapper = FastMCPAuthWrapper(
-            mcp_server=app,
-            oauth_provider=oauth_provider,
-            policy_engine=policy_engine
-        )
-        
-        # Store wrapper for reference
-        app._auth_wrapper = auth_wrapper
-        
-        logger.info("Tool authentication enabled successfully")
-        
-    except Exception as e:
-        logger.error(f"Failed to enable tool authentication: {e}")
-
-
-def cleanup_auth_sessions(app: FastMCP):
-    """Clean up expired auth sessions and tokens"""
-    oauth_provider = get_oauth_provider(app)
-    if oauth_provider:
-        oauth_provider.cleanup_expired_sessions()
-        logger.debug("Cleaned up expired OAuth sessions")

mcp_auth_http_adapter.py

@@ -1,383 +0,0 @@
-"""
-HTTP adapter for MCP Auth Toolkit OAuth endpoints
-Exposes MCP OAuth tools as HTTP endpoints for Claude.ai integration
-"""
-
-import os
-import logging
-import secrets
-import time
-from typing import Optional
-from urllib.parse import urlencode, quote
-from datetime import datetime, timedelta
-
-from fastapi import APIRouter, Request, Query, HTTPException
-from fastapi.responses import RedirectResponse, JSONResponse
-
-# Try to import Clerk SDK
-try:
-    from clerk_backend_api import Clerk
-    CLERK_AVAILABLE = True
-except ImportError as e:
-    CLERK_AVAILABLE = False
-    Clerk = None
-
-logger = logging.getLogger(__name__)
-
-router = APIRouter()
-
-# OAuth configuration
-BASE_URL = os.getenv("BASE_URL", "https://yargimcp.com")
-
-
-@router.get("/.well-known/oauth-authorization-server")
-async def get_oauth_metadata():
-    """OAuth 2.0 Authorization Server Metadata (RFC 8414)"""
-    return JSONResponse({
-        "issuer": BASE_URL,
-        "authorization_endpoint": f"{BASE_URL}/authorize",
-        "token_endpoint": f"{BASE_URL}/token",
-        "registration_endpoint": f"{BASE_URL}/register",
-        "response_types_supported": ["code"],
-        "grant_types_supported": ["authorization_code", "refresh_token"],
-        "code_challenge_methods_supported": ["S256"],
-        "token_endpoint_auth_methods_supported": ["none"],
-        "scopes_supported": ["mcp:tools:read", "mcp:tools:write", "openid", "profile", "email"],
-        "service_documentation": f"{BASE_URL}/mcp/"
-    })
-
-
-@router.get("/.well-known/oauth-protected-resource")
-async def get_protected_resource_metadata():
-    """OAuth Protected Resource Metadata (RFC 9728)"""
-    return JSONResponse({
-        "resource": BASE_URL,
-        "authorization_servers": [BASE_URL],
-        "bearer_methods_supported": ["header"],
-        "scopes_supported": ["mcp:tools:read", "mcp:tools:write"],
-        "resource_documentation": f"{BASE_URL}/docs"
-    })
-
-
-@router.get("/authorize")
-async def authorize_endpoint(
-    response_type: str = Query(...),
-    client_id: str = Query(...),
-    redirect_uri: str = Query(...),
-    code_challenge: str = Query(...),
-    code_challenge_method: str = Query("S256"),
-    state: Optional[str] = Query(None),
-    scope: Optional[str] = Query(None)
-):
-    """OAuth 2.1 Authorization Endpoint - Uses Clerk SDK for custom domains"""
-    
-    logger.info(f"OAuth authorize request - client_id: {client_id}, redirect_uri: {redirect_uri}")
-    
-    if not CLERK_AVAILABLE:
-        logger.error("Clerk SDK not available")
-        raise HTTPException(status_code=500, detail="Clerk SDK not available")
-    
-    # Store OAuth session for later validation
-    try:
-        from mcp_server_main import app as mcp_app
-        from mcp_auth_factory import get_oauth_provider
-        
-        oauth_provider = get_oauth_provider(mcp_app)
-        if not oauth_provider:
-            raise HTTPException(status_code=500, detail="OAuth provider not configured")
-        
-        # Generate session and store PKCE
-        session_id = secrets.token_urlsafe(32)
-        if state is None:
-            state = secrets.token_urlsafe(16)
-        
-        # Create PKCE challenge
-        from mcp_auth.oauth import PKCEChallenge
-        pkce = PKCEChallenge()
-        
-        # Store session data
-        session_data = {
-            "pkce_verifier": pkce.verifier,
-            "pkce_challenge": code_challenge,  # Store the client's challenge
-            "state": state,
-            "redirect_uri": redirect_uri,
-            "client_id": client_id,
-            "scopes": scope.split(" ") if scope else ["mcp:tools:read", "mcp:tools:write"],
-            "created_at": time.time(),
-            "expires_at": (datetime.utcnow() + timedelta(minutes=10)).timestamp(),
-        }
-        oauth_provider.storage.set_session(session_id, session_data)
-        
-        # For Clerk with custom domains, we need to use their hosted sign-in page
-        # We'll pass our callback URL and session info in the state
-        callback_url = f"{BASE_URL}/auth/callback"
-        
-        # Encode session info in state for retrieval after Clerk auth
-        combined_state = f"{state}:{session_id}"
-        
-        # Use Clerk's sign-in URL with proper parameters
-        clerk_domain = os.getenv("CLERK_DOMAIN", "accounts.yargimcp.com")
-        sign_in_params = {
-            "redirect_url": f"{callback_url}?state={quote(combined_state)}",
-        }
-        
-        sign_in_url = f"https://{clerk_domain}/sign-in?{urlencode(sign_in_params)}"
-        
-        logger.info(f"Redirecting to Clerk sign-in: {sign_in_url}")
-        
-        return RedirectResponse(url=sign_in_url)
-        
-    except Exception as e:
-        logger.exception(f"Authorization failed: {e}")
-        raise HTTPException(status_code=500, detail=str(e))
-
-
-@router.get("/auth/callback")
-async def oauth_callback(
-    request: Request,
-    state: Optional[str] = Query(None),
-    clerk_token: Optional[str] = Query(None)
-):
-    """Handle OAuth callback from Clerk - supports both JWT token and cookie auth"""
-    
-    logger.info(f"OAuth callback received - state: {state}")
-    logger.info(f"Query params: {dict(request.query_params)}")
-    logger.info(f"Cookies: {dict(request.cookies)}")
-    logger.info(f"Clerk JWT token provided: {bool(clerk_token)}")
-    
-    # Support both JWT token (for cross-domain) and cookie auth (for subdomain)
-    
-    try:
-        if not state:
-            logger.error("No state parameter provided")
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_request", "error_description": "Missing state parameter"}
-            )
-        
-        # Parse state to get original state and session ID
-        try:
-            if ":" in state:
-                original_state, session_id = state.rsplit(":", 1)
-            else:
-                original_state = state
-                session_id = state  # Fallback
-        except ValueError:
-            logger.error(f"Invalid state format: {state}")
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_request", "error_description": "Invalid state format"}
-            )
-        
-        # Get OAuth provider
-        from mcp_server_main import app as mcp_app
-        from mcp_auth_factory import get_oauth_provider
-        
-        oauth_provider = get_oauth_provider(mcp_app)
-        if not oauth_provider:
-            raise HTTPException(status_code=500, detail="OAuth provider not configured")
-        
-        # Get stored session
-        oauth_session = oauth_provider.storage.get_session(session_id)
-        
-        if not oauth_session:
-            logger.error(f"OAuth session not found for ID: {session_id}")
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_request", "error_description": "OAuth session expired or not found"}
-            )
-        
-        # Check if we have a JWT token (for cross-domain auth)
-        user_authenticated = False
-        auth_method = "none"
-        
-        if clerk_token:
-            logger.info("Attempting JWT token validation")
-            try:
-                # Validate JWT token with Clerk
-                from clerk_backend_api import Clerk
-                clerk = Clerk(bearer_auth=os.getenv("CLERK_SECRET_KEY"))
-                
-                # Extract session_id from JWT token and verify with Clerk
-                import jwt
-                decoded_token = jwt.decode(clerk_token, options={"verify_signature": False})
-                session_id = decoded_token.get("sid") or decoded_token.get("session_id")
-                
-                if session_id:
-                    # Verify with Clerk using session_id
-                    session = clerk.sessions.verify(session_id=session_id, token=clerk_token)
-                    user_id = session.user_id if session else None
-                else:
-                    user_id = None
-                
-                if user_id:
-                    logger.info(f"JWT token validation successful - user_id: {user_id}")
-                    user_authenticated = True
-                    auth_method = "jwt_token"
-                    # Store user info in session for token exchange
-                    oauth_session["user_id"] = user_id
-                    oauth_session["auth_method"] = "jwt_token"
-                else:
-                    logger.error("JWT token validation failed - no user_id in claims")
-            except Exception as e:
-                logger.error(f"JWT token validation failed: {str(e)}")
-                # Fall through to cookie validation
-        
-        # If no JWT token or validation failed, check cookies
-        if not user_authenticated:
-            logger.info("Checking for Clerk session cookies")
-            # Check for Clerk session cookies (for subdomain auth)
-            clerk_session_cookie = request.cookies.get("__session")
-            if clerk_session_cookie:
-                logger.info("Found Clerk session cookie, assuming authenticated")
-                user_authenticated = True
-                auth_method = "cookie"
-                oauth_session["auth_method"] = "cookie"
-            else:
-                logger.info("No Clerk session cookie found")
-        
-        # For custom domains, we'll also trust that Clerk redirected here
-        if not user_authenticated:
-            logger.info("Trusting Clerk redirect for custom domain flow")
-            user_authenticated = True
-            auth_method = "trusted_redirect"
-            oauth_session["auth_method"] = "trusted_redirect"
-        
-        logger.info(f"User authenticated: {user_authenticated}, method: {auth_method}")
-        
-        # Generate simple authorization code for custom domain flow
-        auth_code = f"clerk_custom_{session_id}_{int(time.time())}"
-        
-        # Store the code mapping for token exchange  
-        code_data = {
-            "session_id": session_id,
-            "clerk_authenticated": user_authenticated,
-            "auth_method": auth_method,
-            "custom_domain_flow": True,
-            "created_at": time.time(),
-            "expires_at": (datetime.utcnow() + timedelta(minutes=5)).timestamp(),
-        }
-        if "user_id" in oauth_session:
-            code_data["user_id"] = oauth_session["user_id"]
-            
-        oauth_provider.storage.set_session(f"code_{auth_code}", code_data)
-        
-        # Build redirect URL back to Claude
-        redirect_params = {
-            "code": auth_code,
-            "state": original_state
-        }
-        
-        redirect_url = f"{oauth_session['redirect_uri']}?{urlencode(redirect_params)}"
-        logger.info(f"Redirecting back to Claude: {redirect_url}")
-        
-        return RedirectResponse(url=redirect_url)
-        
-    except Exception as e:
-        logger.exception(f"Callback processing failed: {e}")
-        return JSONResponse(
-            status_code=500,
-            content={"error": "server_error", "error_description": str(e)}
-        )
-
-
-@router.post("/register")
-async def register_client(request: Request):
-    """Dynamic Client Registration (RFC 7591)"""
-    
-    data = await request.json()
-    logger.info(f"Client registration request: {data}")
-    
-    # Simple dynamic registration - accept any client
-    client_id = f"mcp-client-{os.urandom(8).hex()}"
-    
-    return JSONResponse({
-        "client_id": client_id,
-        "client_secret": None,  # Public client
-        "redirect_uris": data.get("redirect_uris", []),
-        "grant_types": ["authorization_code", "refresh_token"],
-        "response_types": ["code"],
-        "client_name": data.get("client_name", "MCP Client"),
-        "token_endpoint_auth_method": "none",
-        "client_id_issued_at": int(datetime.now().timestamp())
-    })
-
-
-@router.post("/token")
-async def token_endpoint(request: Request):
-    """OAuth 2.1 Token Endpoint"""
-    
-    # Parse form data
-    form_data = await request.form()
-    grant_type = form_data.get("grant_type")
-    code = form_data.get("code")
-    redirect_uri = form_data.get("redirect_uri")
-    client_id = form_data.get("client_id")
-    code_verifier = form_data.get("code_verifier")
-    
-    logger.info(f"Token exchange - grant_type: {grant_type}, code: {code[:20] if code else 'None'}...")
-    
-    if grant_type != "authorization_code":
-        return JSONResponse(
-            status_code=400,
-            content={"error": "unsupported_grant_type"}
-        )
-    
-    try:
-        # OAuth token exchange - validate code and return Clerk JWT
-        # This supports proper OAuth flow while using Clerk JWT tokens
-        
-        if not code or not redirect_uri:
-            logger.error("Missing required parameters: code or redirect_uri")
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_request", "error_description": "Missing code or redirect_uri"}
-            )
-        
-        # Validate OAuth code with Clerk
-        if CLERK_AVAILABLE:
-            try:
-                clerk = Clerk(bearer_auth=os.getenv("CLERK_SECRET_KEY"))
-                
-                # In a real implementation, you'd validate the code with Clerk
-                # For now, we'll assume the code is valid if it looks like a Clerk code
-                if len(code) > 10:  # Basic validation
-                    # Create a mock session with the code
-                    # In practice, this would be validated with Clerk's OAuth flow
-                    
-                    # Return Clerk JWT token format
-                    # This should be the actual Clerk JWT token from the OAuth flow
-                    return JSONResponse({
-                        "access_token": f"mock_clerk_jwt_{code}",
-                        "token_type": "Bearer",
-                        "expires_in": 3600,
-                        "scope": "yargi.read yargi.search"
-                    })
-                else:
-                    logger.error(f"Invalid code format: {code}")
-                    return JSONResponse(
-                        status_code=400,
-                        content={"error": "invalid_grant", "error_description": "Invalid authorization code"}
-                    )
-                    
-            except Exception as e:
-                logger.error(f"Clerk validation failed: {e}")
-                return JSONResponse(
-                    status_code=400,
-                    content={"error": "invalid_grant", "error_description": "Authorization code validation failed"}
-                )
-        else:
-            logger.warning("Clerk SDK not available, using mock response")
-            return JSONResponse({
-                "access_token": "mock_jwt_token_for_development",
-                "token_type": "Bearer",
-                "expires_in": 3600,
-                "scope": "yargi.read yargi.search"
-            })
-        
-    except Exception as e:
-        logger.exception(f"Token exchange failed: {e}")
-        return JSONResponse(
-            status_code=500,
-            content={"error": "server_error", "error_description": str(e)}
-        )

mcp_auth_http_simple.py

@@ -1,522 +0,0 @@
-"""
-Simplified MCP OAuth HTTP adapter - only Clerk JWT based authentication
-Uses Redis for authorization code storage to support multi-machine deployment
-"""
-
-import os
-import logging
-from typing import Optional
-from urllib.parse import urlencode, quote
-
-from fastapi import APIRouter, Request, Query, HTTPException
-from fastapi.responses import RedirectResponse, JSONResponse
-
-# Import Redis session store
-from redis_session_store import get_redis_store
-
-# Try to import Clerk SDK
-try:
-    from clerk_backend_api import Clerk
-    CLERK_AVAILABLE = True
-except ImportError:
-    CLERK_AVAILABLE = False
-    Clerk = None
-
-logger = logging.getLogger(__name__)
-
-router = APIRouter()
-
-# OAuth configuration
-BASE_URL = os.getenv("BASE_URL", "https://api.yargimcp.com")
-CLERK_DOMAIN = os.getenv("CLERK_DOMAIN", "accounts.yargimcp.com")
-
-# Initialize Redis store
-redis_store = None
-
-def get_redis_session_store():
-    """Get Redis store instance with lazy initialization."""
-    global redis_store
-    if redis_store is None:
-        try:
-            import concurrent.futures
-            import functools
-            
-            # Use thread pool with timeout to prevent hanging
-            with concurrent.futures.ThreadPoolExecutor(max_workers=1) as executor:
-                future = executor.submit(get_redis_store)
-                try:
-                    # 5 second timeout for Redis initialization
-                    redis_store = future.result(timeout=5.0)
-                    if redis_store:
-                        logger.info("Redis session store initialized for OAuth handler")
-                    else:
-                        logger.warning("Redis store initialization returned None")
-                except concurrent.futures.TimeoutError:
-                    logger.error("Redis initialization timed out after 5 seconds")
-                    redis_store = None
-                    future.cancel()  # Try to cancel the hanging operation
-                    
-        except Exception as e:
-            logger.error(f"Failed to initialize Redis store: {e}")
-            redis_store = None
-            
-        if redis_store is None:
-            # Fall back to in-memory storage with warning
-            logger.warning("Falling back to in-memory storage - multi-machine deployment will not work")
-            
-    return redis_store
-
-@router.get("/.well-known/oauth-authorization-server")
-async def get_oauth_metadata():
-    """OAuth 2.0 Authorization Server Metadata (RFC 8414)"""
-    return JSONResponse({
-        "issuer": BASE_URL,
-        "authorization_endpoint": "https://yargimcp.com/mcp-callback",
-        "token_endpoint": f"{BASE_URL}/token",
-        "registration_endpoint": f"{BASE_URL}/register",
-        "response_types_supported": ["code"],
-        "grant_types_supported": ["authorization_code"],
-        "code_challenge_methods_supported": ["S256"],
-        "token_endpoint_auth_methods_supported": ["none"],
-        "scopes_supported": ["read", "search", "openid", "profile", "email"],
-        "service_documentation": f"{BASE_URL}/mcp/"
-    })
-
-@router.get("/auth/login")
-async def oauth_authorize(
-    request: Request,
-    client_id: str = Query(...),
-    redirect_uri: str = Query(...),
-    response_type: str = Query("code"),
-    scope: Optional[str] = Query("read search"),
-    state: Optional[str] = Query(None),
-    code_challenge: Optional[str] = Query(None),
-    code_challenge_method: Optional[str] = Query(None)
-):
-    """OAuth 2.1 Authorization Endpoint - redirects to Clerk"""
-    
-    logger.info(f"OAuth authorize request - client_id: {client_id}")
-    logger.info(f"Redirect URI: {redirect_uri}")
-    logger.info(f"State: {state}")
-    logger.info(f"PKCE Challenge: {bool(code_challenge)}")
-    
-    try:
-        # Build callback URL with all necessary parameters
-        callback_url = f"{BASE_URL}/auth/callback"
-        callback_params = {
-            "client_id": client_id,
-            "redirect_uri": redirect_uri,
-            "state": state or "",
-            "scope": scope or "read search"
-        }
-        
-        # Add PKCE parameters if present
-        if code_challenge:
-            callback_params["code_challenge"] = code_challenge
-            callback_params["code_challenge_method"] = code_challenge_method or "S256"
-        
-        # Encode callback URL as redirect_url for Clerk
-        callback_with_params = f"{callback_url}?{urlencode(callback_params)}"
-        
-        # Build Clerk sign-in URL - use yargimcp.com frontend for JWT token generation
-        clerk_params = {
-            "redirect_url": callback_with_params
-        }
-        
-        # Use frontend sign-in page that handles JWT token generation
-        clerk_signin_url = f"https://yargimcp.com/sign-in?{urlencode(clerk_params)}"
-        
-        logger.info(f"Redirecting to Clerk: {clerk_signin_url}")
-        
-        return RedirectResponse(url=clerk_signin_url)
-        
-    except Exception as e:
-        logger.exception(f"Authorization failed: {e}")
-        raise HTTPException(status_code=500, detail=str(e))
-
-@router.get("/auth/callback")
-async def oauth_callback(
-    request: Request,
-    client_id: str = Query(...),
-    redirect_uri: str = Query(...),
-    state: Optional[str] = Query(None),
-    scope: Optional[str] = Query("read search"),
-    code_challenge: Optional[str] = Query(None),
-    code_challenge_method: Optional[str] = Query(None),
-    clerk_token: Optional[str] = Query(None)
-):
-    """OAuth callback from Clerk - generates authorization code"""
-    
-    logger.info(f"OAuth callback - client_id: {client_id}")
-    logger.info(f"Clerk token provided: {bool(clerk_token)}")
-    
-    try:
-        # Validate user with Clerk and generate real JWT token
-        user_authenticated = False
-        user_id = None
-        session_id = None
-        real_jwt_token = None
-        
-        if clerk_token and CLERK_AVAILABLE:
-            try:
-                # Extract user info from JWT token (no Clerk session verification needed)
-                import jwt
-                decoded_token = jwt.decode(clerk_token, options={"verify_signature": False})
-                user_id = decoded_token.get("user_id") or decoded_token.get("sub")
-                user_email = decoded_token.get("email")
-                token_scopes = decoded_token.get("scopes", ["read", "search"])
-                
-                logger.info(f"JWT token claims - user_id: {user_id}, email: {user_email}, scopes: {token_scopes}")
-                
-                if user_id and user_email:
-                    # JWT token is already signed by Clerk and contains valid user info
-                    user_authenticated = True
-                    logger.info(f"User authenticated via JWT token - user_id: {user_id}")
-                    
-                    # Use the JWT token directly as the real token (it's already from Clerk template)
-                    real_jwt_token = clerk_token
-                    logger.info("Using Clerk JWT token directly (already real token)")
-                    
-                else:
-                    logger.error(f"Missing required fields in JWT token - user_id: {bool(user_id)}, email: {bool(user_email)}")
-                    
-            except Exception as e:
-                logger.error(f"JWT validation failed: {e}")
-        
-        # Fallback to cookie validation
-        if not user_authenticated:
-            clerk_session = request.cookies.get("__session")
-            if clerk_session:
-                user_authenticated = True
-                logger.info("User authenticated via cookie")
-                
-                # Try to get session from cookie and generate JWT
-                if CLERK_AVAILABLE:
-                    try:
-                        clerk = Clerk(bearer_auth=os.getenv("CLERK_SECRET_KEY"))
-                        # Note: sessions.verify_session is deprecated, but we'll try
-                        # In practice, you'd need to extract session_id from cookie
-                        logger.info("Cookie authentication - JWT generation not implemented yet")
-                    except Exception as e:
-                        logger.warning(f"Failed to generate JWT from cookie: {e}")
-        
-        # Only generate authorization code if we have a real JWT token
-        if user_authenticated and real_jwt_token:
-            # Generate authorization code
-            auth_code = f"clerk_auth_{os.urandom(16).hex()}"
-            
-            # Prepare code data
-            import time
-            code_data = {
-                "user_id": user_id,
-                "session_id": session_id,
-                "real_jwt_token": real_jwt_token,
-                "user_authenticated": user_authenticated,
-                "client_id": client_id,
-                "redirect_uri": redirect_uri,
-                "scope": scope or "read search"
-            }
-            
-            # Try to store in Redis, fall back to in-memory if Redis unavailable
-            store = get_redis_session_store()
-            if store:
-                # Store in Redis with automatic expiration
-                success = store.set_oauth_code(auth_code, code_data)
-                if success:
-                    logger.info(f"Stored authorization code {auth_code[:10]}... in Redis with real JWT token")
-                else:
-                    logger.error(f"Failed to store authorization code in Redis, falling back to in-memory")
-                    # Fall back to in-memory storage
-                    if not hasattr(oauth_callback, '_code_storage'):
-                        oauth_callback._code_storage = {}
-                    oauth_callback._code_storage[auth_code] = code_data
-            else:
-                # Fall back to in-memory storage
-                logger.warning("Redis not available, using in-memory storage")
-                if not hasattr(oauth_callback, '_code_storage'):
-                    oauth_callback._code_storage = {}
-                oauth_callback._code_storage[auth_code] = code_data
-                logger.info(f"Stored authorization code in memory (fallback)")
-            
-            # Redirect back to client with authorization code
-            redirect_params = {
-                "code": auth_code,
-                "state": state or ""
-            }
-            
-            final_redirect_url = f"{redirect_uri}?{urlencode(redirect_params)}"
-            logger.info(f"Redirecting back to client: {final_redirect_url}")
-            
-            return RedirectResponse(url=final_redirect_url)
-        else:
-            # No JWT token yet - redirect back to sign-in page to wait for authentication
-            logger.info("No JWT token provided - redirecting back to sign-in to complete authentication")
-            
-            # Keep the same redirect URL so the flow continues
-            sign_in_params = {
-                "redirect_url": f"{request.url._url}"  # Current callback URL with all params
-            }
-            
-            sign_in_url = f"https://yargimcp.com/sign-in?{urlencode(sign_in_params)}"
-            logger.info(f"Redirecting back to sign-in: {sign_in_url}")
-            
-            return RedirectResponse(url=sign_in_url)
-        
-    except Exception as e:
-        logger.exception(f"Callback processing failed: {e}")
-        return JSONResponse(
-            status_code=500,
-            content={"error": "server_error", "error_description": str(e)}
-        )
-
-@router.post("/auth/register")
-async def register_client(request: Request):
-    """Dynamic Client Registration (RFC 7591)"""
-    
-    data = await request.json()
-    logger.info(f"Client registration request: {data}")
-    
-    # Simple dynamic registration - accept any client
-    client_id = f"mcp-client-{os.urandom(8).hex()}"
-    
-    return JSONResponse({
-        "client_id": client_id,
-        "client_secret": None,  # Public client
-        "redirect_uris": data.get("redirect_uris", []),
-        "grant_types": ["authorization_code"],
-        "response_types": ["code"],
-        "client_name": data.get("client_name", "MCP Client"),
-        "token_endpoint_auth_method": "none"
-    })
-
-@router.post("/auth/callback")
-async def oauth_callback_post(request: Request):
-    """OAuth callback POST endpoint for token exchange"""
-    
-    # Parse form data (standard OAuth token exchange format)
-    form_data = await request.form()
-    grant_type = form_data.get("grant_type")
-    code = form_data.get("code")
-    redirect_uri = form_data.get("redirect_uri")
-    client_id = form_data.get("client_id")
-    code_verifier = form_data.get("code_verifier")
-    
-    logger.info(f"OAuth callback POST - grant_type: {grant_type}")
-    logger.info(f"Code: {code[:20] if code else 'None'}...")
-    logger.info(f"Client ID: {client_id}")
-    logger.info(f"PKCE verifier: {bool(code_verifier)}")
-    
-    if grant_type != "authorization_code":
-        return JSONResponse(
-            status_code=400,
-            content={"error": "unsupported_grant_type"}
-        )
-    
-    if not code or not redirect_uri:
-        return JSONResponse(
-            status_code=400,
-            content={"error": "invalid_request", "error_description": "Missing code or redirect_uri"}
-        )
-    
-    try:
-        # Validate authorization code
-        if not code.startswith("clerk_auth_"):
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_grant", "error_description": "Invalid authorization code"}
-            )
-        
-        # Retrieve stored JWT token using authorization code from Redis or in-memory fallback
-        stored_code_data = None
-        
-        # Try to get from Redis first, then fall back to in-memory
-        store = get_redis_session_store()
-        if store:
-            stored_code_data = store.get_oauth_code(code, delete_after_use=True)
-            if stored_code_data:
-                logger.info(f"Retrieved authorization code {code[:10]}... from Redis")
-            else:
-                logger.warning(f"Authorization code {code[:10]}... not found in Redis")
-        
-        # Fall back to in-memory storage if Redis unavailable or code not found
-        if not stored_code_data and hasattr(oauth_callback, '_code_storage'):
-            stored_code_data = oauth_callback._code_storage.get(code)
-            if stored_code_data:
-                # Clean up in-memory storage
-                oauth_callback._code_storage.pop(code, None)
-                logger.info(f"Retrieved authorization code {code[:10]}... from in-memory storage")
-        
-        if not stored_code_data:
-            logger.error(f"No stored data found for authorization code: {code}")
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_grant", "error_description": "Authorization code not found or expired"}
-            )
-        
-        # Note: Redis TTL handles expiration automatically, but check for manual expiration for in-memory fallback
-        import time
-        expires_at = stored_code_data.get("expires_at", 0)
-        if expires_at and time.time() > expires_at:
-            logger.error(f"Authorization code expired: {code}")
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_grant", "error_description": "Authorization code expired"}
-            )
-        
-        # Get the real JWT token
-        real_jwt_token = stored_code_data.get("real_jwt_token")
-        
-        if real_jwt_token:
-            logger.info("Returning real Clerk JWT token")
-            # Note: Code already deleted from Redis, clean up in-memory fallback if used
-            if hasattr(oauth_callback, '_code_storage'):
-                oauth_callback._code_storage.pop(code, None)
-            
-            return JSONResponse({
-                "access_token": real_jwt_token,
-                "token_type": "Bearer",
-                "expires_in": 3600,
-                "scope": "read search"
-            })
-        else:
-            logger.warning("No real JWT token found, generating mock token")
-            # Fallback to mock token for testing
-            mock_token = f"mock_clerk_jwt_{code}"
-            return JSONResponse({
-                "access_token": mock_token,
-                "token_type": "Bearer",
-                "expires_in": 3600,
-                "scope": "read search"
-            })
-        
-    except Exception as e:
-        logger.exception(f"OAuth callback POST failed: {e}")
-        return JSONResponse(
-            status_code=500,
-            content={"error": "server_error", "error_description": str(e)}
-        )
-
-@router.post("/register")
-async def register_client(request: Request):
-    """Dynamic Client Registration (RFC 7591)"""
-    
-    data = await request.json()
-    logger.info(f"Client registration request: {data}")
-    
-    # Simple dynamic registration - accept any client
-    client_id = f"mcp-client-{os.urandom(8).hex()}"
-    
-    return JSONResponse({
-        "client_id": client_id,
-        "client_secret": None,  # Public client
-        "redirect_uris": data.get("redirect_uris", []),
-        "grant_types": ["authorization_code"],
-        "response_types": ["code"],
-        "client_name": data.get("client_name", "MCP Client"),
-        "token_endpoint_auth_method": "none"
-    })
-
-@router.post("/token")
-async def token_endpoint(request: Request):
-    """OAuth 2.1 Token Endpoint - exchanges code for Clerk JWT"""
-    
-    # Parse form data
-    form_data = await request.form()
-    grant_type = form_data.get("grant_type")
-    code = form_data.get("code")
-    redirect_uri = form_data.get("redirect_uri")
-    client_id = form_data.get("client_id")
-    code_verifier = form_data.get("code_verifier")
-    
-    logger.info(f"Token exchange - grant_type: {grant_type}")
-    logger.info(f"Code: {code[:20] if code else 'None'}...")
-    
-    if grant_type != "authorization_code":
-        return JSONResponse(
-            status_code=400,
-            content={"error": "unsupported_grant_type"}
-        )
-    
-    if not code or not redirect_uri:
-        return JSONResponse(
-            status_code=400,
-            content={"error": "invalid_request", "error_description": "Missing code or redirect_uri"}
-        )
-    
-    try:
-        # Validate authorization code
-        if not code.startswith("clerk_auth_"):
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_grant", "error_description": "Invalid authorization code"}
-            )
-        
-        # Retrieve stored JWT token using authorization code from Redis or in-memory fallback
-        stored_code_data = None
-        
-        # Try to get from Redis first, then fall back to in-memory
-        store = get_redis_session_store()
-        if store:
-            stored_code_data = store.get_oauth_code(code, delete_after_use=True)
-            if stored_code_data:
-                logger.info(f"Retrieved authorization code {code[:10]}... from Redis (/token endpoint)")
-            else:
-                logger.warning(f"Authorization code {code[:10]}... not found in Redis (/token endpoint)")
-        
-        # Fall back to in-memory storage if Redis unavailable or code not found
-        if not stored_code_data and hasattr(oauth_callback, '_code_storage'):
-            stored_code_data = oauth_callback._code_storage.get(code)
-            if stored_code_data:
-                # Clean up in-memory storage
-                oauth_callback._code_storage.pop(code, None)
-                logger.info(f"Retrieved authorization code {code[:10]}... from in-memory storage (/token endpoint)")
-        
-        if not stored_code_data:
-            logger.error(f"No stored data found for authorization code: {code}")
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_grant", "error_description": "Authorization code not found or expired"}
-            )
-        
-        # Note: Redis TTL handles expiration automatically, but check for manual expiration for in-memory fallback
-        import time
-        expires_at = stored_code_data.get("expires_at", 0)
-        if expires_at and time.time() > expires_at:
-            logger.error(f"Authorization code expired: {code}")
-            return JSONResponse(
-                status_code=400,
-                content={"error": "invalid_grant", "error_description": "Authorization code expired"}
-            )
-        
-        # Get the real JWT token
-        real_jwt_token = stored_code_data.get("real_jwt_token")
-        
-        if real_jwt_token:
-            logger.info("Returning real Clerk JWT token from /token endpoint")
-            # Note: Code already deleted from Redis, clean up in-memory fallback if used
-            if hasattr(oauth_callback, '_code_storage'):
-                oauth_callback._code_storage.pop(code, None)
-            
-            return JSONResponse({
-                "access_token": real_jwt_token,
-                "token_type": "Bearer",
-                "expires_in": 3600,
-                "scope": "read search"
-            })
-        else:
-            logger.warning("No real JWT token found in /token endpoint, generating mock token")
-            # Fallback to mock token for testing
-            mock_token = f"mock_clerk_jwt_{code}"
-            return JSONResponse({
-                "access_token": mock_token,
-                "token_type": "Bearer",
-                "expires_in": 3600,
-                "scope": "read search"
-            })
-        
-    except Exception as e:
-        logger.exception(f"Token exchange failed: {e}")
-        return JSONResponse(
-            status_code=500,
-            content={"error": "server_error", "error_description": str(e)}
-        )

migration_app.py

@@ -0,0 +1,57 @@
+"""
+Migration stub for the deprecated Yargı MCP endpoint.
+
+Exposes a single tool that informs the MCP client the server has moved
+and the user must update their configuration.
+
+Entrypoint variable `app` is a FastMCP instance so it works with
+Dokploy's FastMCP build pipeline (`fastmcp inspect`, `fastmcp run`).
+
+Run with:
+    fastmcp run migration_app.py:app --transport http --port 8000
+"""
+
+from starlette.responses import JSONResponse
+from fastmcp import FastMCP
+
+NEW_URL = "https://yargimcp.surucu.dev/mcp"
+
+app = FastMCP(
+    name="Yargı MCP (taşındı / moved)",
+    instructions=(
+        f"Bu Yargı MCP endpoint'i kullanımdan kaldırıldı. "
+        f"Sunucu yeni adrese taşındı: {NEW_URL}. "
+        f"Kullanıcıya MCP istemci ayarlarındaki sunucu URL'sini güncellemesini söyle."
+    ),
+)
+
+
+@app.tool(
+    description=(
+        "DEPRECATED ENDPOINT — Yargı MCP sunucusu yeni adrese taşındı. "
+        "Bu endpoint'teki eski araçlar (Yargıtay, Danıştay, Anayasa Mahkemesi, "
+        "KVKK, BDDK, Rekabet Kurumu, Sayıştay, KİK, Emsal, Uyuşmazlık, GİB, "
+        "Sigorta Tahkim, Bedesten vb.) artık burada çalışmıyor. "
+        "Bu aracı çağırarak yeni sunucu adresini ve migrasyon talimatını al, "
+        "ardından kullanıcıya MCP istemci (Claude Desktop, 5ire, ChatGPT vb.) "
+        "ayarlarındaki sunucu URL'sini güncellemesi gerektiğini açıkça söyle."
+    )
+)
+def migration_notice() -> dict:
+    """Yargı MCP sunucusunun yeni adresini ve migrasyon talimatını döner."""
+    return {
+        "status": "moved",
+        "new_url": NEW_URL,
+        "message": (
+            f"Yargı MCP sunucusu yeni adrese taşındı: {NEW_URL}\n\n"
+            f"Lütfen MCP istemcinin (Claude Desktop, 5ire, ChatGPT vb.) "
+            f"ayarlarındaki sunucu URL'sini yukarıdaki yeni adresle güncelleyin. "
+            f"Mevcut endpoint artık kullanım dışıdır ve sadece bu uyarıyı döner."
+        ),
+    }
+
+
+@app.custom_route("/health", methods=["GET"])
+async def health(request):
+    """Health check endpoint for monitoring services."""
+    return JSONResponse({"status": "deprecated", "new_url": NEW_URL})

nginx.conf

@@ -1,94 +0,0 @@
-events {
-    worker_connections 1024;
-}
-
-http {
-    upstream yargi_mcp {
-        server yargi-mcp:8000;
-    }
-
-    # Rate limiting
-    limit_req_zone $binary_remote_addr zone=api_limit:10m rate=10r/s;
-    limit_req_zone $binary_remote_addr zone=mcp_limit:10m rate=100r/s;
-
-    server {
-        listen 80;
-        server_name localhost;
-
-        # Redirect HTTP to HTTPS in production
-        # return 301 https://$server_name$request_uri;
-
-        # Security headers
-        add_header X-Content-Type-Options nosniff;
-        add_header X-Frame-Options DENY;
-        add_header X-XSS-Protection "1; mode=block";
-        add_header Referrer-Policy "strict-origin-when-cross-origin";
-
-        # API endpoints
-        location /api/ {
-            limit_req zone=api_limit burst=20 nodelay;
-            
-            proxy_pass http://yargi_mcp;
-            proxy_set_header Host $host;
-            proxy_set_header X-Real-IP $remote_addr;
-            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
-            proxy_set_header X-Forwarded-Proto $scheme;
-            
-            # Timeouts
-            proxy_connect_timeout 60s;
-            proxy_send_timeout 60s;
-            proxy_read_timeout 60s;
-        }
-
-        # MCP endpoint (higher rate limit)
-        location /mcp-server/mcp/ {
-            limit_req zone=mcp_limit burst=50 nodelay;
-            
-            proxy_pass http://yargi_mcp;
-            proxy_set_header Host $host;
-            proxy_set_header X-Real-IP $remote_addr;
-            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
-            proxy_set_header X-Forwarded-Proto $scheme;
-            
-            # WebSocket support
-            proxy_http_version 1.1;
-            proxy_set_header Upgrade $http_upgrade;
-            proxy_set_header Connection "upgrade";
-            
-            # Longer timeouts for MCP operations
-            proxy_connect_timeout 300s;
-            proxy_send_timeout 300s;
-            proxy_read_timeout 300s;
-        }
-
-        # Health check (no rate limit)
-        location /health {
-            proxy_pass http://yargi_mcp;
-            proxy_set_header Host $host;
-        }
-
-        # Root and other paths
-        location / {
-            limit_req zone=api_limit burst=10 nodelay;
-            
-            proxy_pass http://yargi_mcp;
-            proxy_set_header Host $host;
-            proxy_set_header X-Real-IP $remote_addr;
-            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
-            proxy_set_header X-Forwarded-Proto $scheme;
-        }
-    }
-
-    # SSL configuration (uncomment for production)
-    # server {
-    #     listen 443 ssl http2;
-    #     server_name your-domain.com;
-    #
-    #     ssl_certificate /etc/nginx/ssl/cert.pem;
-    #     ssl_certificate_key /etc/nginx/ssl/key.pem;
-    #     ssl_protocols TLSv1.2 TLSv1.3;
-    #     ssl_ciphers HIGH:!aNULL:!MD5;
-    #
-    #     # Include all location blocks from above
-    # }
-}

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "yargi-mcp"
-version = "0.1.6"
+version = "0.2.1"
 description = "MCP Server For Turkish Legal Databases"
 readme = "README.md"
 requires-python = ">=3.11"
@@ -25,12 +25,12 @@ dependencies = [
     "markitdown[pdf]>=0.1.1",
     "pydantic>=2.11.4",
     "aiohttp>=3.11.18",
-    "playwright>=1.52.0",
     "fastmcp>=2.10.5",
     "pypdf>=5.5.0",
     "fastapi>=0.115.14",
-    "PyJWT>=2.8.0",
-    "tiktoken>=0.5.0",
+    "cryptography>=44.0.0",
+    "openai>=1.0.0",
+    "numpy>=1.24.0",
 ]
 
 [project.optional-dependencies]
@@ -46,20 +46,15 @@ production = [
     "gunicorn>=22.0.0",
     "uvicorn[standard]>=0.30.0",
 ]
-saas = [
-    "clerk-backend-api>=3.0.0",
-    "stripe>=9.1.0",
-    "upstash-redis>=1.1.0",
-]
 
 [project.scripts]
 yargi-mcp = "mcp_server_main:main"
 
 [tool.setuptools]
-py-modules = ["mcp_server_main", "mcp_auth_factory", "mcp_auth_http_adapter", "asgi_app", "fastapi_app", "starlette_app", "run_asgi", "stripe_webhook"]
+py-modules = ["mcp_server_main", "asgi_app"]
 
 [tool.setuptools.packages.find]
-include = ["*_mcp_module", "mcp_auth"]
+include = ["*_mcp_module", "semantic_search"]
 
 [build-system]
 requires = ["setuptools>=65.0", "wheel"]

rekabet_mcp_module/client.py

@@ -1,5 +1,6 @@
 # rekabet_mcp_module/client.py
 
+import asyncio
 import httpx
 from bs4 import BeautifulSoup
 from typing import List, Optional, Tuple, Dict, Any
@@ -141,12 +142,12 @@ class RekabetKurumuApiClient:
 
                     # Row 1: Publication Date, Decision Number, Related Cases Link
                     td_elements_r1 = rows[0].find_all("td")
-                    pub_date = td_elements_r1[0].get_text(strip=True) if len(td_elements_r1) > 0 else None
-                    dec_num = td_elements_r1[1].get_text(strip=True) if len(td_elements_r1) > 1 else None
+                    pub_date = td_elements_r1[0].get_text(strip=True) if len(td_elements_r1) > 0 else ""
+                    dec_num = td_elements_r1[1].get_text(strip=True) if len(td_elements_r1) > 1 else ""
 
                     related_cases_link_tag = td_elements_r1[2].find("a", href=True) if len(td_elements_r1) > 2 else None
-                    related_cases_url_str: Optional[str] = None
-                    karar_id_from_related: Optional[str] = None
+                    related_cases_url_str: str = ""
+                    karar_id_from_related: str = ""
                     if related_cases_link_tag and related_cases_link_tag.has_attr('href'):
                         related_cases_url_str = urljoin(self.BASE_URL, related_cases_link_tag['href'])
                         qs_related = parse_qs(urlparse(related_cases_link_tag['href']).query)
@@ -155,16 +156,16 @@ class RekabetKurumuApiClient:
 
                     # Row 2: Decision Date, Decision Type
                     td_elements_r2 = rows[1].find_all("td")
-                    dec_date = td_elements_r2[0].get_text(strip=True) if len(td_elements_r2) > 0 else None
-                    dec_type_text = td_elements_r2[1].get_text(strip=True) if len(td_elements_r2) > 1 else None
+                    dec_date = td_elements_r2[0].get_text(strip=True) if len(td_elements_r2) > 0 else ""
+                    dec_type_text = td_elements_r2[1].get_text(strip=True) if len(td_elements_r2) > 1 else ""
 
                     # Row 3: Title and Main Decision Link
                     title_cell = rows[2].find("td", colspan="5")
                     decision_link_tag = title_cell.find("a", href=True) if title_cell else None
 
-                    title_text: Optional[str] = None
-                    decision_landing_url_str: Optional[str] = None 
-                    karar_id_from_main_link: Optional[str] = None
+                    title_text: str = ""
+                    decision_landing_url_str: str = ""
+                    karar_id_from_main_link: str = ""
 
                     if decision_link_tag and decision_link_tag.has_attr('href'):
                         title_text = decision_link_tag.get_text(strip=True)
@@ -185,16 +186,12 @@ class RekabetKurumuApiClient:
                         logger.warning(f"Table {idx+1} Karar ID not found. Skipping. Title (if any): {title_text}")
                         continue
 
-                    # Convert string URLs to HttpUrl for the model
-                    final_decision_url = HttpUrl(decision_landing_url_str) if decision_landing_url_str else None
-                    final_related_cases_url = HttpUrl(related_cases_url_str) if related_cases_url_str else None
-
                     processed_decisions.append(RekabetDecisionSummary(
                         publication_date=pub_date, decision_number=dec_num, decision_date=dec_date,
                         decision_type_text=dec_type_text, title=title_text,
-                        decision_url=final_decision_url, 
+                        decision_url=decision_landing_url_str,
                         karar_id=current_karar_id,
-                        related_cases_url=final_related_cases_url
+                        related_cases_url=related_cases_url_str
                     ))
                     logger.debug(f"Table {idx+1} parsed successfully: Karar ID '{current_karar_id}', Title '{title_text[:50] if title_text else 'N/A'}...'")
 
@@ -357,7 +354,7 @@ class RekabetKurumuApiClient:
                     total_pdf_pages = total_pdf_pages_from_extraction 
 
                     if single_page_pdf_bytes:
-                        markdown_for_requested_page = self._convert_pdf_bytes_to_markdown(single_page_pdf_bytes, str(pdf_url_to_report or full_landing_page_url))
+                        markdown_for_requested_page = await asyncio.to_thread(self._convert_pdf_bytes_to_markdown, single_page_pdf_bytes, str(pdf_url_to_report or full_landing_page_url))
                         if not markdown_for_requested_page:
                             error_message = (error_message or "") + f"; Could not convert page {page_number} of PDF to Markdown."
                     elif total_pdf_pages > 0 : 

requirements.txt

@@ -1,11 +0,0 @@
-fastmcp
-httpx
-beautifulsoup4
-markitdown[pdf]
-pydantic
-aiohttp
-playwright
-pypdf
-fastapi>=0.115.14
-uvicorn[standard]>=0.30.0
-starlette>=0.37.0

run_asgi.py

@@ -1,119 +0,0 @@
-#!/usr/bin/env python3
-"""
-Standalone ASGI server runner for Yargı MCP
-
-This script provides a simple way to run the Yargı MCP server
-as a web service using uvicorn.
-
-Usage:
-    python run_asgi.py
-    python run_asgi.py --host 0.0.0.0 --port 8080
-    python run_asgi.py --reload  # For development
-"""
-
-import os
-import sys
-import argparse
-import logging
-from pathlib import Path
-
-# Add project root to Python path
-sys.path.insert(0, str(Path(__file__).parent))
-
-try:
-    import uvicorn
-except ImportError:
-    print("Error: uvicorn is not installed.")
-    print("Please install it with: pip install uvicorn")
-    sys.exit(1)
-
-# Configure logging
-logging.basicConfig(
-    level=logging.INFO,
-    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
-)
-
-def main():
-    parser = argparse.ArgumentParser(
-        description="Run Yargı MCP server as an ASGI web service"
-    )
-    parser.add_argument(
-        "--host",
-        type=str,
-        default=os.getenv("HOST", "127.0.0.1"),
-        help="Host to bind to (default: 127.0.0.1)"
-    )
-    parser.add_argument(
-        "--port",
-        type=int,
-        default=int(os.getenv("PORT", "8000")),
-        help="Port to bind to (default: 8000)"
-    )
-    parser.add_argument(
-        "--reload",
-        action="store_true",
-        help="Enable auto-reload for development"
-    )
-    parser.add_argument(
-        "--transport",
-        choices=["http", "sse"],
-        default="http",
-        help="Transport type (default: http)"
-    )
-    parser.add_argument(
-        "--log-level",
-        choices=["debug", "info", "warning", "error"],
-        default=os.getenv("LOG_LEVEL", "info").lower(),
-        help="Log level (default: info)"
-    )
-    parser.add_argument(
-        "--workers",
-        type=int,
-        default=1,
-        help="Number of worker processes (default: 1)"
-    )
-    
-    args = parser.parse_args()
-    
-    # Select app based on transport
-    app_name = "asgi_app:app" if args.transport == "http" else "asgi_app:sse_app"
-    
-    # Configure uvicorn
-    config = {
-        "app": app_name,
-        "host": args.host,
-        "port": args.port,
-        "log_level": args.log_level,
-        "reload": args.reload,
-        "access_log": True,
-    }
-    
-    # Add workers only if not in reload mode
-    if not args.reload and args.workers > 1:
-        config["workers"] = args.workers
-    
-    # Print startup information
-    print(f"Starting Yargı MCP server...")
-    print(f"Host: {args.host}")
-    print(f"Port: {args.port}")
-    print(f"Transport: {args.transport}")
-    print(f"Log level: {args.log_level}")
-    if args.reload:
-        print("Auto-reload: enabled")
-    else:
-        print(f"Workers: {args.workers}")
-    print(f"\nServer will be available at: http://{args.host}:{args.port}")
-    print(f"MCP endpoint: http://{args.host}:{args.port}/mcp/")
-    print(f"Health check: http://{args.host}:{args.port}/health")
-    print(f"API status: http://{args.host}:{args.port}/status")
-    print("\nPress CTRL+C to stop the server\n")
-    
-    # Run uvicorn
-    try:
-        uvicorn.run(**config)
-    except KeyboardInterrupt:
-        print("\nShutting down server...")
-        sys.exit(0)
-
-if __name__ == "__main__":
-    main()

sayistay_mcp_module/client.py

@@ -1,5 +1,6 @@
 # sayistay_mcp_module/client.py
 
+import asyncio
 import httpx
 import re
 from bs4 import BeautifulSoup
@@ -48,6 +49,12 @@ class SayistayApiClient:
     TEMYIZ_KURULU_ENDPOINT = "/KararlarTemyiz/DataTablesList"
     DAIRE_ENDPOINT = "/KararlarDaire/DataTablesList"
 
+    # Marker present in the upstream WAF block page (also returns HTTP 418).
+    # Verified 2026-05-03 against real Chrome — the block targets POSTs to
+    # the DataTablesList endpoints regardless of headers/cookies/CSRF, so
+    # we surface a specific error instead of the generic "I'm a teapot".
+    _WAF_BLOCK_MARKER = "Bilgi Güvenliği Politikaları Gereği Kısıtlanmıştır"
+    
     # Page endpoints for session initialization and document access
     GENEL_KURUL_PAGE = "/KararlarGenelKurul"
     TEMYIZ_KURULU_PAGE = "/KararlarTemyiz"
@@ -141,6 +148,23 @@ class SayistayApiClient:
         
         return enum_value
 
+    def _raise_if_waf_blocked(self, response: httpx.Response, endpoint_label: str) -> None:
+        """
+        Sayıştay's upstream WAF returns HTTP 418 with a Turkish HTML block
+        page for POSTs to the DataTablesList endpoints. This affects every
+        client (verified with real Chrome on 2026-05-03), so there is no
+        client-side workaround. Detect it and raise a clear error.
+        """
+        if response.status_code == 418 or self._WAF_BLOCK_MARKER in response.text:
+            raise RuntimeError(
+                f"Sayıştay upstream WAF blocked the {endpoint_label} request "
+                f"(HTTP {response.status_code} from {response.request.url}). "
+                "This is a server-side restriction at sayistay.gov.tr — affects "
+                "all clients including a real browser — and cannot be worked "
+                "around from yargi-mcp. Try again later or contact Sayıştay if "
+                "the block persists."
+            )
+
     def _build_datatables_params(self, start: int, length: int, draw: int = 1) -> List[Tuple[str, str]]:
         """Build standard DataTables parameters for all endpoints."""
         params = [
@@ -384,6 +408,7 @@ class SayistayApiClient:
                 data=encoded_data,
                 headers=headers
             )
+            self._raise_if_waf_blocked(response, "Genel Kurul")
             response.raise_for_status()
             response_json = response.json()
 
@@ -443,6 +468,7 @@ class SayistayApiClient:
                 data=encoded_data,
                 headers=headers
             )
+            self._raise_if_waf_blocked(response, "Temyiz Kurulu")
             response.raise_for_status()
             response_json = response.json()
             
@@ -502,6 +528,7 @@ class SayistayApiClient:
                 data=encoded_data,
                 headers=headers
             )
+            self._raise_if_waf_blocked(response, "Daire")
             response.raise_for_status()
             response_json = response.json()
             
@@ -631,7 +658,7 @@ class SayistayApiClient:
                 )
             
             # Convert HTML to Markdown using existing method
-            markdown_content = self._convert_html_to_markdown(html_content)
+            markdown_content = await asyncio.to_thread(self._convert_html_to_markdown, html_content)
             
             if markdown_content and "Error converting HTML content" not in markdown_content:
                 logger.info(f"Successfully retrieved and converted document {decision_id} to Markdown")

semantic_search/__init__.py

@@ -0,0 +1,23 @@
+# semantic_search/__init__.py
+
+from .embedder import (
+    OpenRouterEmbedder,
+    LocalEmbedder,
+    get_embedder,
+    is_openrouter_available,
+    is_local_embedding_configured,
+    is_semantic_search_available,
+)
+from .vector_store import VectorStore
+from .processor import DocumentProcessor
+
+__all__ = [
+    'OpenRouterEmbedder',
+    'LocalEmbedder',
+    'get_embedder',
+    'is_openrouter_available',
+    'is_local_embedding_configured',
+    'is_semantic_search_available',
+    'VectorStore',
+    'DocumentProcessor',
+]

semantic_search/embedder.py

@@ -0,0 +1,348 @@
+# semantic_search/embedder.py
+
+import logging
+import os
+from typing import Dict, List, Optional
+import numpy as np
+
+logger = logging.getLogger(__name__)
+
+
+# OpenRouter defaults (preserve backward compatibility)
+DEFAULT_MODEL = "google/gemini-embedding-001"
+DEFAULT_DIMENSION = 3072
+
+# Local provider defaults — Ollama with nomic-embed-text out of the box.
+# Override via LOCAL_EMBEDDING_BASE_URL / LOCAL_EMBEDDING_MODEL /
+# LOCAL_EMBEDDING_DIMENSION when using a different server or model.
+# For Turkish, intfloat/multilingual-e5-large (1024 dims, prompt_style=e5)
+# served via HuggingFace TEI is the recommended setup — see README.
+LOCAL_DEFAULT_BASE_URL = "http://localhost:11434/v1"
+LOCAL_DEFAULT_MODEL = "nomic-embed-text"
+LOCAL_DEFAULT_DIMENSION = 768
+
+# Prompt-template styles. Embedding models are trained with specific
+# prefixes — using the wrong style silently degrades retrieval quality.
+#   - "gemini": "task: {task} | query: {text}" / "title: {title} | text: {text}"
+#               (matches google/gemini-embedding-001, the OpenRouter default)
+#   - "e5":     "query: {text}" / "passage: {text}"
+#               (matches intfloat/multilingual-e5-* models — best for Turkish)
+#   - "raw":    no prefix; pass text through as-is
+PROMPT_STYLES = ("gemini", "e5", "raw")
+DEFAULT_PROMPT_STYLE = "gemini"
+
+
+def _format_query(prompt_style: str, query: str, task: str) -> str:
+    if prompt_style == "e5":
+        return f"query: {query}"
+    if prompt_style == "raw":
+        return query
+    # gemini (default)
+    return f"task: {task} | query: {query}"
+
+
+def _format_document(prompt_style: str, doc: str, title: str) -> str:
+    if prompt_style == "e5":
+        return f"passage: {doc}"
+    if prompt_style == "raw":
+        return doc
+    # gemini (default)
+    return f"title: {title} | text: {doc}"
+
+
+def _resolve_prompt_style(explicit: Optional[str], default: str) -> str:
+    style = (explicit or os.getenv("EMBEDDING_PROMPT_STYLE") or default).strip().lower()
+    if style not in PROMPT_STYLES:
+        raise ValueError(
+            f"Unknown EMBEDDING_PROMPT_STYLE {style!r}; expected one of {PROMPT_STYLES}"
+        )
+    return style
+
+
+def is_openrouter_available() -> bool:
+    """Check if OpenRouter API key is available."""
+    return bool(os.getenv("OPENROUTER_API_KEY"))
+
+
+def is_local_embedding_configured() -> bool:
+    """Check if the user opted into a local embedding endpoint."""
+    return os.getenv("EMBEDDING_PROVIDER", "").strip().lower() == "local"
+
+
+def is_semantic_search_available() -> bool:
+    """Returns True if any embedding provider is configured."""
+    return is_local_embedding_configured() or is_openrouter_available()
+
+
+def _coerce_dimension(value, env_name: str, default: int) -> int:
+    """Parse a dimension value (int or str) with clear error messages."""
+    if value is None:
+        return default
+    try:
+        parsed = int(value)
+    except (TypeError, ValueError) as e:
+        raise ValueError(
+            f"{env_name} must be an integer, got {value!r}"
+        ) from e
+    if parsed <= 0:
+        raise ValueError(f"Embedding dimension must be positive, got {parsed}")
+    return parsed
+
+
+class _BaseOpenAICompatibleEmbedder:
+    """
+    Shared encode/similarity logic for embedders backed by the OpenAI Python
+    SDK. Subclasses configure ``client``, ``model``, ``dimension``, and
+    optionally ``_extra_headers`` (e.g. OpenRouter ranking headers).
+    """
+
+    # Subclasses may override; sent on every embeddings.create call when set.
+    _extra_headers: Dict[str, str] = {}
+
+    # Set by subclasses
+    client = None
+    model: str = ""
+    dimension: int = 0
+    prompt_style: str = DEFAULT_PROMPT_STYLE
+
+    def encode_query(self, query: str, task: str = "search result") -> np.ndarray:
+        """
+        Encode a search query. Prefix is selected by ``self.prompt_style``.
+
+        Args:
+            query: The search query text
+            task: Task hint used by the gemini-style prefix; ignored for
+                e5/raw styles.
+
+        Returns:
+            Numpy array of embeddings (``self.dimension`` elements).
+        """
+        text = _format_query(self.prompt_style, query, task)
+
+        try:
+            response = self.client.embeddings.create(
+                model=self.model,
+                input=text,
+                encoding_format="float",
+                extra_headers=self._extra_headers or None,
+            )
+
+            embedding = np.array(response.data[0].embedding, dtype=np.float32)
+
+            # L2 normalize for cosine similarity
+            norm = np.linalg.norm(embedding)
+            if norm > 0:
+                embedding = embedding / norm
+
+            logger.debug(f"Encoded query: {query[:50]}... -> shape: {embedding.shape}")
+            return embedding
+
+        except Exception as e:
+            logger.error(f"Failed to encode query: {e}")
+            raise
+
+    def encode_documents(self, documents: List[str], titles: Optional[List[str]] = None) -> np.ndarray:
+        """
+        Encode multiple documents with a batch API call.
+
+        Args:
+            documents: List of document texts
+            titles: Optional list of document titles
+
+        Returns:
+            Numpy array of embeddings (N x ``self.dimension``).
+        """
+        if not documents:
+            return np.array([])
+
+        texts = []
+        for i, doc in enumerate(documents):
+            title = titles[i] if titles and i < len(titles) else "none"
+            texts.append(_format_document(self.prompt_style, doc, title))
+
+        try:
+            response = self.client.embeddings.create(
+                model=self.model,
+                input=texts,
+                encoding_format="float",
+                extra_headers=self._extra_headers or None,
+            )
+
+            embeddings = np.array(
+                [d.embedding for d in sorted(response.data, key=lambda x: x.index)],
+                dtype=np.float32,
+            )
+
+            # L2 normalize each embedding for cosine similarity
+            norms = np.linalg.norm(embeddings, axis=1, keepdims=True)
+            embeddings = embeddings / (norms + 1e-8)
+
+            logger.info(f"Encoded {len(documents)} documents -> shape: {embeddings.shape}")
+            return embeddings
+
+        except Exception as e:
+            logger.error(f"Failed to encode documents: {e}")
+            raise
+
+    def compute_similarity(self, query_embedding: np.ndarray, document_embeddings: np.ndarray) -> np.ndarray:
+        """
+        Compute cosine similarity between query and documents.
+
+        Args:
+            query_embedding: Query embedding (``self.dimension``,)
+            document_embeddings: Document embeddings (N x ``self.dimension``)
+
+        Returns:
+            Similarity scores (N,)
+        """
+        if len(query_embedding.shape) == 1:
+            query_embedding = query_embedding.reshape(1, -1)
+
+        # Embeddings are already L2-normalized.
+        similarities = np.dot(document_embeddings, query_embedding.T).squeeze()
+        return similarities
+
+
+class OpenRouterEmbedder(_BaseOpenAICompatibleEmbedder):
+    """
+    Embedder using OpenRouter's embedding API.
+
+    The model and dimension are configurable so users can pick any OpenRouter
+    embedding model (e.g. when one becomes paid). Configuration precedence:
+    explicit constructor args > environment variables > defaults.
+
+    Environment variables:
+        OPENROUTER_API_KEY (required): OpenRouter credential
+        OPENROUTER_EMBEDDING_MODEL (optional): override the embedding model id
+        OPENROUTER_EMBEDDING_DIMENSION (optional): override the vector size
+
+    Defaults preserve backward compatibility: ``google/gemini-embedding-001``
+    at 3072 dimensions.
+    """
+
+    _extra_headers = {
+        "HTTP-Referer": "https://yargimcp.com",
+        "X-Title": "Yargi MCP Server",
+    }
+
+    def __init__(
+        self,
+        model: Optional[str] = None,
+        dimension: Optional[int] = None,
+        prompt_style: Optional[str] = None,
+    ):
+        api_key = os.getenv("OPENROUTER_API_KEY")
+        if not api_key:
+            raise ValueError("OPENROUTER_API_KEY environment variable is not set")
+
+        try:
+            from openai import OpenAI
+        except ImportError:
+            raise ImportError("openai package is required. Install with: pip install openai")
+
+        self.client = OpenAI(
+            base_url="https://openrouter.ai/api/v1",
+            api_key=api_key,
+        )
+        self.model = model or os.getenv("OPENROUTER_EMBEDDING_MODEL") or DEFAULT_MODEL
+        self.dimension = _coerce_dimension(
+            dimension if dimension is not None else os.getenv("OPENROUTER_EMBEDDING_DIMENSION"),
+            "OPENROUTER_EMBEDDING_DIMENSION",
+            DEFAULT_DIMENSION,
+        )
+        # Default to gemini-style prefix for OpenRouter — matches the default
+        # google/gemini-embedding-001 model. Override via constructor or
+        # EMBEDDING_PROMPT_STYLE env var when picking a different model.
+        self.prompt_style = _resolve_prompt_style(prompt_style, "gemini")
+
+        logger.info(
+            f"OpenRouter Embedder initialized with model: {self.model} "
+            f"(dimension={self.dimension}, prompt_style={self.prompt_style})"
+        )
+
+
+class LocalEmbedder(_BaseOpenAICompatibleEmbedder):
+    """
+    Embedder for a local OpenAI-compatible embedding server — Ollama,
+    llama.cpp, vLLM, LM Studio, etc. Zero new Python dependencies; just
+    point the existing OpenAI SDK at a local base URL.
+
+    Environment variables:
+        EMBEDDING_PROVIDER=local              (selects this provider)
+        LOCAL_EMBEDDING_BASE_URL              (default: http://localhost:11434/v1)
+        LOCAL_EMBEDDING_MODEL                 (default: nomic-embed-text)
+        LOCAL_EMBEDDING_DIMENSION             (default: 768)
+        LOCAL_EMBEDDING_API_KEY               (optional; ignored by most local servers)
+
+    Setup (Ollama):
+        $ ollama serve
+        $ ollama pull nomic-embed-text          # or bge-m3 for better Turkish
+
+    The dimension MUST match the model's actual output size (e.g. 768 for
+    nomic-embed-text, 1024 for bge-m3, 1024 for mxbai-embed-large).
+    """
+
+    def __init__(
+        self,
+        base_url: Optional[str] = None,
+        model: Optional[str] = None,
+        dimension: Optional[int] = None,
+        api_key: Optional[str] = None,
+        prompt_style: Optional[str] = None,
+    ):
+        try:
+            from openai import OpenAI
+        except ImportError:
+            raise ImportError("openai package is required. Install with: pip install openai")
+
+        self.base_url = (
+            base_url
+            or os.getenv("LOCAL_EMBEDDING_BASE_URL")
+            or LOCAL_DEFAULT_BASE_URL
+        )
+        # Most local servers don't validate the key — use a placeholder so
+        # the OpenAI SDK doesn't error on the missing-key check.
+        effective_key = (
+            api_key
+            or os.getenv("LOCAL_EMBEDDING_API_KEY")
+            or "no-key-needed"
+        )
+
+        self.client = OpenAI(base_url=self.base_url, api_key=effective_key)
+        self.model = model or os.getenv("LOCAL_EMBEDDING_MODEL") or LOCAL_DEFAULT_MODEL
+        self.dimension = _coerce_dimension(
+            dimension if dimension is not None else os.getenv("LOCAL_EMBEDDING_DIMENSION"),
+            "LOCAL_EMBEDDING_DIMENSION",
+            LOCAL_DEFAULT_DIMENSION,
+        )
+        # Default to e5 prefix for local — the recommended Turkish setup
+        # (multilingual-e5-large). Override via EMBEDDING_PROMPT_STYLE when
+        # using a different model family (e.g. nomic, bge).
+        self.prompt_style = _resolve_prompt_style(prompt_style, "e5")
+
+        logger.info(
+            f"Local Embedder initialized: model={self.model} "
+            f"base_url={self.base_url} dimension={self.dimension} "
+            f"prompt_style={self.prompt_style}"
+        )
+
+
+def get_embedder():
+    """
+    Factory that picks the embedder based on EMBEDDING_PROVIDER.
+
+    - ``EMBEDDING_PROVIDER=local`` -> ``LocalEmbedder``
+    - otherwise -> ``OpenRouterEmbedder`` (requires OPENROUTER_API_KEY)
+
+    Raises:
+        ValueError: If no provider is configured (neither local nor OpenRouter).
+    """
+    if is_local_embedding_configured():
+        return LocalEmbedder()
+    if is_openrouter_available():
+        return OpenRouterEmbedder()
+    raise ValueError(
+        "No embedding provider configured. Set OPENROUTER_API_KEY for hosted "
+        "embeddings, or EMBEDDING_PROVIDER=local (with LOCAL_EMBEDDING_* "
+        "env vars) for a local OpenAI-compatible server like Ollama."
+    )

semantic_search/processor.py

@@ -0,0 +1,305 @@
+# semantic_search/processor.py
+
+import logging
+import re
+from typing import List, Dict, Any, Optional
+from dataclasses import dataclass
+import hashlib
+
+logger = logging.getLogger(__name__)
+
+@dataclass
+class DocumentChunk:
+    """Represents a chunk of a document."""
+    chunk_id: str
+    document_id: str
+    text: str
+    metadata: Dict[str, Any]
+    chunk_index: int
+    total_chunks: int
+
+class DocumentProcessor:
+    """
+    Processes legal documents for semantic search.
+    Handles chunking, cleaning, and metadata extraction.
+    """
+    
+    def __init__(self, 
+                 chunk_size: int = 1000,
+                 chunk_overlap: int = 200,
+                 min_chunk_size: int = 100):
+        """
+        Initialize document processor.
+        
+        Args:
+            chunk_size: Target size for each chunk in characters
+            chunk_overlap: Number of overlapping characters between chunks
+            min_chunk_size: Minimum chunk size to keep
+        """
+        self.chunk_size = chunk_size
+        self.chunk_overlap = chunk_overlap
+        self.min_chunk_size = min_chunk_size
+        
+        logger.info(f"Initialized DocumentProcessor (chunk_size={chunk_size}, overlap={chunk_overlap})")
+    
+    def process_document(self, 
+                        document_id: str,
+                        text: str,
+                        metadata: Optional[Dict[str, Any]] = None) -> List[DocumentChunk]:
+        """
+        Process a single document into chunks.
+        
+        Args:
+            document_id: Unique document identifier
+            text: Document text content
+            metadata: Optional document metadata
+            
+        Returns:
+            List of document chunks
+        """
+        if not text or len(text.strip()) < self.min_chunk_size:
+            logger.warning(f"Document {document_id} too short to process")
+            return []
+        
+        # Clean text
+        cleaned_text = self._clean_text(text)
+        
+        # Extract metadata from text if not provided
+        if metadata is None:
+            metadata = {}
+        
+        # Add extracted metadata
+        extracted_metadata = self._extract_metadata(cleaned_text)
+        metadata.update(extracted_metadata)
+        
+        # Create chunks
+        chunks = self._create_chunks(cleaned_text)
+        
+        # Create DocumentChunk objects
+        document_chunks = []
+        for i, chunk_text in enumerate(chunks):
+            chunk_id = self._generate_chunk_id(document_id, i)
+            
+            chunk = DocumentChunk(
+                chunk_id=chunk_id,
+                document_id=document_id,
+                text=chunk_text,
+                metadata={
+                    **metadata,
+                    'chunk_index': i,
+                    'total_chunks': len(chunks)
+                },
+                chunk_index=i,
+                total_chunks=len(chunks)
+            )
+            document_chunks.append(chunk)
+        
+        logger.info(f"Processed document {document_id} into {len(chunks)} chunks")
+        return document_chunks
+    
+    def _clean_text(self, text: str) -> str:
+        """
+        Clean and normalize text for processing.
+        
+        Args:
+            text: Raw text
+            
+        Returns:
+            Cleaned text
+        """
+        # Remove excessive whitespace
+        text = re.sub(r'\s+', ' ', text)
+        
+        # Remove special characters but keep Turkish characters
+        # Keep: letters, numbers, spaces, and common punctuation
+        text = re.sub(r'[^\w\s\.\,\;\:\!\?\-\(\)\"\'ÇĞIİÖŞÜçğıiöşü]', ' ', text)
+        
+        # Remove multiple spaces
+        text = re.sub(r' +', ' ', text)
+        
+        # Trim
+        text = text.strip()
+        
+        return text
+    
+    def _extract_metadata(self, text: str) -> Dict[str, Any]:
+        """
+        Extract metadata from legal document text.
+        
+        Args:
+            text: Document text
+            
+        Returns:
+            Extracted metadata
+        """
+        metadata = {}
+        
+        # Extract case numbers (Esas/Karar)
+        esas_pattern = r'E(?:sas)?[\s\.\:]*(\d{4})[\/\-](\d+)'
+        karar_pattern = r'K(?:arar)?[\s\.\:]*(\d{4})[\/\-](\d+)'
+        
+        esas_match = re.search(esas_pattern, text[:500])  # Look in first 500 chars
+        if esas_match:
+            metadata['esas_no'] = f"E.{esas_match.group(1)}/{esas_match.group(2)}"
+        
+        karar_match = re.search(karar_pattern, text[:500])
+        if karar_match:
+            metadata['karar_no'] = f"K.{karar_match.group(1)}/{karar_match.group(2)}"
+        
+        # Extract dates (DD.MM.YYYY or DD/MM/YYYY format)
+        date_pattern = r'(\d{1,2})[\.\/](\d{1,2})[\.\/](\d{4})'
+        dates = re.findall(date_pattern, text[:1000])  # Look in first 1000 chars
+        if dates:
+            # Take the first date as decision date
+            day, month, year = dates[0]
+            metadata['karar_tarihi'] = f"{year}-{month.zfill(2)}-{day.zfill(2)}"
+        
+        # Extract court/chamber name
+        chamber_patterns = [
+            r'(\d+)\.\s*Hukuk\s+Dairesi',
+            r'(\d+)\.\s*Ceza\s+Dairesi',
+            r'Hukuk\s+Genel\s+Kurulu',
+            r'Ceza\s+Genel\s+Kurulu',
+            r'(\d+)\.\s*Daire'
+        ]
+        
+        for pattern in chamber_patterns:
+            match = re.search(pattern, text[:500], re.IGNORECASE)
+            if match:
+                metadata['chamber'] = match.group(0)
+                break
+        
+        return metadata
+    
+    def _create_chunks(self, text: str) -> List[str]:
+        """
+        Create overlapping chunks from text.
+        
+        Args:
+            text: Cleaned document text
+            
+        Returns:
+            List of text chunks
+        """
+        chunks = []
+        
+        # Split by sentences for better semantic coherence
+        sentences = self._split_sentences(text)
+        
+        current_chunk = []
+        current_size = 0
+        
+        for sentence in sentences:
+            sentence_size = len(sentence)
+            
+            # If adding this sentence exceeds chunk size
+            if current_size + sentence_size > self.chunk_size and current_chunk:
+                # Save current chunk
+                chunk_text = ' '.join(current_chunk)
+                chunks.append(chunk_text)
+                
+                # Create overlap for next chunk
+                overlap_size = 0
+                overlap_sentences = []
+                
+                # Add sentences from the end until we reach overlap size
+                for sent in reversed(current_chunk):
+                    overlap_size += len(sent)
+                    overlap_sentences.insert(0, sent)
+                    if overlap_size >= self.chunk_overlap:
+                        break
+                
+                # Start new chunk with overlap
+                current_chunk = overlap_sentences
+                current_size = sum(len(s) for s in current_chunk)
+            
+            # Add sentence to current chunk
+            current_chunk.append(sentence)
+            current_size += sentence_size
+        
+        # Add final chunk if not empty
+        if current_chunk:
+            chunk_text = ' '.join(current_chunk)
+            if len(chunk_text) >= self.min_chunk_size:
+                chunks.append(chunk_text)
+        
+        return chunks
+    
+    def _split_sentences(self, text: str) -> List[str]:
+        """
+        Split text into sentences.
+        
+        Args:
+            text: Text to split
+            
+        Returns:
+            List of sentences
+        """
+        # Simple sentence splitting for Turkish text
+        # Split on period, question mark, exclamation, but not on abbreviations
+        
+        # Common Turkish abbreviations to preserve
+        abbreviations = ['Dr', 'Prof', 'Av', 'Md', 'Yrd', 'Doç', 'No', 'S', 'vs', 'vb', 'bkz']
+        
+        # Replace abbreviations temporarily
+        temp_text = text
+        replacements = {}
+        for i, abbr in enumerate(abbreviations):
+            placeholder = f"__ABBR{i}__"
+            temp_text = temp_text.replace(f"{abbr}.", placeholder)
+            replacements[placeholder] = f"{abbr}."
+        
+        # Split sentences
+        sentence_endings = re.compile(r'[.!?]+')
+        sentences = sentence_endings.split(temp_text)
+        
+        # Restore abbreviations and clean
+        cleaned_sentences = []
+        for sentence in sentences:
+            # Restore abbreviations
+            for placeholder, original in replacements.items():
+                sentence = sentence.replace(placeholder, original)
+            
+            # Clean and add if not empty
+            sentence = sentence.strip()
+            if sentence and len(sentence) > 10:  # Minimum sentence length
+                cleaned_sentences.append(sentence)
+        
+        return cleaned_sentences
+    
+    def _generate_chunk_id(self, document_id: str, chunk_index: int) -> str:
+        """
+        Generate unique chunk ID.
+        
+        Args:
+            document_id: Parent document ID
+            chunk_index: Index of chunk in document
+            
+        Returns:
+            Unique chunk ID
+        """
+        chunk_string = f"{document_id}_chunk_{chunk_index}"
+        chunk_hash = hashlib.md5(chunk_string.encode()).hexdigest()[:8]
+        return f"{document_id}_c{chunk_index}_{chunk_hash}"
+    
+    def combine_chunks(self, chunks: List[DocumentChunk]) -> str:
+        """
+        Combine chunks back into full document text.
+        
+        Args:
+            chunks: List of document chunks
+            
+        Returns:
+            Combined text
+        """
+        if not chunks:
+            return ""
+        
+        # Sort by chunk index
+        sorted_chunks = sorted(chunks, key=lambda x: x.chunk_index)
+        
+        # For overlapping chunks, we need to be careful about duplication
+        # Simple approach: just concatenate with space
+        combined = " ".join([chunk.text for chunk in sorted_chunks])
+        
+        return combined

semantic_search/vector_store.py

@@ -0,0 +1,235 @@
+# semantic_search/vector_store.py
+
+import logging
+import numpy as np
+from typing import List, Dict, Any, Tuple, Optional
+from dataclasses import dataclass
+import json
+
+logger = logging.getLogger(__name__)
+
+@dataclass
+class Document:
+    """Represents a document with its embedding and metadata."""
+    id: str
+    text: str
+    embedding: np.ndarray
+    metadata: Dict[str, Any]
+    
+    def to_dict(self) -> Dict[str, Any]:
+        """Convert to dictionary (excluding embedding for serialization)."""
+        return {
+            'id': self.id,
+            'text': self.text,
+            'metadata': self.metadata
+        }
+
+class VectorStore:
+    """
+    In-memory vector storage with similarity search capabilities.
+    Future versions can use Faiss, ChromaDB, or other vector databases.
+    """
+    
+    def __init__(self, dimension: int = 768):
+        """
+        Initialize vector store.
+        
+        Args:
+            dimension: Embedding dimension size
+        """
+        self.dimension = dimension
+        self.documents: List[Document] = []
+        self.embeddings: Optional[np.ndarray] = None
+        self.index_built = False
+        
+        logger.info(f"Initialized VectorStore with dimension: {dimension}")
+    
+    def add_documents(self, 
+                     ids: List[str],
+                     texts: List[str],
+                     embeddings: np.ndarray,
+                     metadata: Optional[List[Dict[str, Any]]] = None) -> int:
+        """
+        Add documents to the vector store.
+        
+        Args:
+            ids: Document IDs
+            texts: Document texts
+            embeddings: Document embeddings (N x dimension)
+            metadata: Optional metadata for each document
+            
+        Returns:
+            Number of documents added
+        """
+        if len(ids) != len(texts) or len(ids) != embeddings.shape[0]:
+            raise ValueError("Mismatched lengths for ids, texts, and embeddings")
+        
+        if metadata and len(metadata) != len(ids):
+            raise ValueError("Metadata length doesn't match document count")
+        
+        # Add documents
+        for i in range(len(ids)):
+            doc = Document(
+                id=ids[i],
+                text=texts[i],
+                embedding=embeddings[i],
+                metadata=metadata[i] if metadata else {}
+            )
+            self.documents.append(doc)
+        
+        # Rebuild index
+        self._build_index()
+        
+        logger.info(f"Added {len(ids)} documents to vector store. Total: {len(self.documents)}")
+        return len(ids)
+    
+    def _build_index(self):
+        """Build or rebuild the embedding index."""
+        if not self.documents:
+            self.embeddings = None
+            self.index_built = False
+            return
+        
+        # Stack all embeddings into a single array
+        self.embeddings = np.vstack([doc.embedding for doc in self.documents])
+        self.index_built = True
+        
+        logger.debug(f"Built index with shape: {self.embeddings.shape}")
+    
+    def search(self, 
+              query_embedding: np.ndarray,
+              top_k: int = 10,
+              threshold: Optional[float] = None) -> List[Tuple[Document, float]]:
+        """
+        Search for similar documents using cosine similarity.
+        
+        Args:
+            query_embedding: Query embedding vector
+            top_k: Number of results to return
+            threshold: Optional similarity threshold (0-1)
+            
+        Returns:
+            List of (Document, similarity_score) tuples
+        """
+        if not self.index_built or self.embeddings is None:
+            logger.warning("No documents in vector store")
+            return []
+        
+        # Ensure query is 2D
+        if len(query_embedding.shape) == 1:
+            query_embedding = query_embedding.reshape(1, -1)
+        
+        # Compute cosine similarities (assuming normalized embeddings)
+        similarities = np.dot(self.embeddings, query_embedding.T).squeeze()
+        
+        # Apply threshold if specified
+        if threshold is not None:
+            valid_indices = np.where(similarities >= threshold)[0]
+            if len(valid_indices) == 0:
+                logger.info(f"No documents above threshold {threshold}")
+                return []
+            similarities = similarities[valid_indices]
+            valid_docs = [self.documents[i] for i in valid_indices]
+        else:
+            valid_docs = self.documents
+        
+        # Get top-k indices
+        top_k = min(top_k, len(valid_docs))
+        if top_k == 0:
+            return []
+        
+        # Use argpartition for efficiency with large arrays
+        if len(similarities) > top_k:
+            top_indices = np.argpartition(similarities, -top_k)[-top_k:]
+            top_indices = top_indices[np.argsort(similarities[top_indices])[::-1]]
+        else:
+            top_indices = np.argsort(similarities)[::-1]
+        
+        # Create results
+        results = []
+        for idx in top_indices:
+            doc = valid_docs[idx] if threshold else self.documents[idx]
+            score = float(similarities[idx])
+            results.append((doc, score))
+        
+        logger.info(f"Search returned {len(results)} results (top_k={top_k})")
+        return results
+    
+    def hybrid_search(self,
+                     query_embedding: np.ndarray,
+                     keyword_scores: Dict[str, float],
+                     top_k: int = 10,
+                     alpha: float = 0.5) -> List[Tuple[Document, float]]:
+        """
+        Hybrid search combining vector similarity and keyword scores.
+        
+        Args:
+            query_embedding: Query embedding vector
+            keyword_scores: Document ID to keyword relevance score mapping
+            top_k: Number of results to return
+            alpha: Weight for vector similarity (1-alpha for keyword score)
+            
+        Returns:
+            List of (Document, combined_score) tuples
+        """
+        if not self.index_built:
+            logger.warning("No documents in vector store")
+            return []
+        
+        # Get vector similarities
+        vector_results = self.search(query_embedding, top_k=len(self.documents))
+        
+        # Combine scores
+        combined_scores = []
+        for doc, vector_score in vector_results:
+            keyword_score = keyword_scores.get(doc.id, 0.0)
+            # Normalize keyword score to 0-1 range if needed
+            if keyword_score > 1.0:
+                keyword_score = keyword_score / max(keyword_scores.values())
+            
+            combined_score = alpha * vector_score + (1 - alpha) * keyword_score
+            combined_scores.append((doc, combined_score))
+        
+        # Sort by combined score and return top-k
+        combined_scores.sort(key=lambda x: x[1], reverse=True)
+        results = combined_scores[:top_k]
+        
+        logger.info(f"Hybrid search returned {len(results)} results")
+        return results
+    
+    def clear(self):
+        """Clear all documents from the store."""
+        self.documents = []
+        self.embeddings = None
+        self.index_built = False
+        logger.info("Cleared vector store")
+    
+    def size(self) -> int:
+        """Get number of documents in store."""
+        return len(self.documents)
+    
+    def get_by_id(self, doc_id: str) -> Optional[Document]:
+        """Get document by ID."""
+        for doc in self.documents:
+            if doc.id == doc_id:
+                return doc
+        return None
+    
+    def get_stats(self) -> Dict[str, Any]:
+        """Get statistics about the vector store."""
+        stats = {
+            'num_documents': len(self.documents),
+            'dimension': self.dimension,
+            'index_built': self.index_built,
+            'memory_usage_mb': 0
+        }
+        
+        if self.embeddings is not None:
+            # Estimate memory usage
+            memory_bytes = self.embeddings.nbytes
+            for doc in self.documents:
+                memory_bytes += len(doc.text.encode('utf-8'))
+                memory_bytes += len(json.dumps(doc.metadata).encode('utf-8'))
+            stats['memory_usage_mb'] = memory_bytes / (1024 * 1024)
+        
+        return stats

sigorta_tahkim_mcp_module/__init__.py

@@ -0,0 +1,21 @@
+# sigorta_tahkim_mcp_module/__init__.py
+
+from .client import SigortaTahkimApiClient
+from .models import (
+    SigortaTahkimSearchRequest,
+    SigortaTahkimDecisionSummary,
+    SigortaTahkimSearchResult,
+    SigortaTahkimDocumentMarkdown,
+    SigortaTahkimSearchWithinMatch,
+    SigortaTahkimSearchWithinResult
+)
+
+__all__ = [
+    "SigortaTahkimApiClient",
+    "SigortaTahkimSearchRequest",
+    "SigortaTahkimDecisionSummary",
+    "SigortaTahkimSearchResult",
+    "SigortaTahkimDocumentMarkdown",
+    "SigortaTahkimSearchWithinMatch",
+    "SigortaTahkimSearchWithinResult"
+]

sigorta_tahkim_mcp_module/client.py

@@ -0,0 +1,345 @@
+# sigorta_tahkim_mcp_module/client.py
+
+import asyncio
+import httpx
+from typing import Optional
+import logging
+import os
+import re
+import io
+import math
+from markitdown import MarkItDown
+
+from .models import (
+    SigortaTahkimSearchRequest,
+    SigortaTahkimDecisionSummary,
+    SigortaTahkimSearchResult,
+    SigortaTahkimDocumentMarkdown,
+    SigortaTahkimSearchWithinMatch,
+    SigortaTahkimSearchWithinResult
+)
+
+logger = logging.getLogger(__name__)
+if not logger.hasHandlers():
+    logging.basicConfig(
+        level=logging.INFO,
+        format='%(asctime)s - %(name)s - %(levelname)s - %(message)s'
+    )
+
+
+# Turkish-specific lowercase: İ→i, I→ı (Python's str.lower() doesn't handle these)
+_TR_UPPER = str.maketrans("İIÇĞÖŞÜ", "iıçğöşü")
+
+
+def _turkish_lower(text: str) -> str:
+    """Lowercase with Turkish İ/I handling."""
+    return text.translate(_TR_UPPER).lower()
+
+
+class SigortaTahkimApiClient:
+    """
+    API client for searching and retrieving Sigorta Tahkim Komisyonu
+    (Insurance Arbitration Commission) decisions using Tavily Search API
+    for discovery and direct PDF download for content retrieval.
+
+    The commission publishes quarterly PDF journals ("Hakem Karar Dergisi")
+    containing arbitration decisions. There are 64 issues spanning 2010-2025.
+    """
+
+    TAVILY_API_URL = "https://api.tavily.com/search"
+    BASE_URL = "https://www.sigortatahkim.org"
+    PDF_BASE_URL = "https://www.sigortatahkim.org/content/CmsFiles/"
+    DOCUMENT_MARKDOWN_CHUNK_SIZE = 5000
+
+    def __init__(self, request_timeout: float = 60.0):
+        """Initialize the Sigorta Tahkim API client."""
+        self.tavily_api_key = os.getenv("TAVILY_API_KEY")
+        if not self.tavily_api_key:
+            self.tavily_api_key = "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"
+            logger.info("Using fallback Tavily API token (development token)")
+        else:
+            logger.info("Using Tavily API key from environment variable")
+
+        self.http_client = httpx.AsyncClient(
+            headers={
+                "User-Agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36"
+            },
+            timeout=httpx.Timeout(request_timeout)
+        )
+        self.markitdown = MarkItDown()
+
+    async def close_client_session(self):
+        """Close the HTTP client session."""
+        await self.http_client.aclose()
+        logger.info("SigortaTahkimApiClient: HTTP client session closed.")
+
+    def _get_pdf_filename(self, issue_number: int) -> str:
+        """Get the PDF filename for a given journal issue number."""
+        if issue_number == 4:
+            return "karardergisisayi4.pdf"
+        elif 57 <= issue_number <= 61:
+            return f"revizekd{issue_number}.pdf"
+        else:
+            return f"karardrgs{issue_number}.pdf"
+
+    def _extract_issue_number(self, url: str) -> Optional[str]:
+        """Extract journal issue number from a sigortatahkim.org URL."""
+        # Pattern: karardrgs{N}.pdf
+        match = re.search(r'karardrgs(\d+)\.pdf', url, re.IGNORECASE)
+        if match:
+            return match.group(1)
+
+        # Pattern: revizekd{N}.pdf
+        match = re.search(r'revizekd(\d+)\.pdf', url, re.IGNORECASE)
+        if match:
+            return match.group(1)
+
+        # Pattern: karardergisisayi{N}.pdf
+        match = re.search(r'karardergisisayi(\d+)\.pdf', url, re.IGNORECASE)
+        if match:
+            return match.group(1)
+
+        # Pattern: sayı or sayi in URL path with number
+        match = re.search(r'say[ıi]\s*[-:]?\s*(\d+)', url, re.IGNORECASE)
+        if match:
+            return match.group(1)
+
+        return None
+
+    async def search_decisions(
+        self,
+        request: SigortaTahkimSearchRequest
+    ) -> SigortaTahkimSearchResult:
+        """
+        Search for Sigorta Tahkim Komisyonu decisions using Tavily API.
+
+        Args:
+            request: Search request parameters
+
+        Returns:
+            SigortaTahkimSearchResult with matching decisions
+        """
+        try:
+            headers = {
+                "Content-Type": "application/json",
+                "Authorization": f"Bearer {self.tavily_api_key}"
+            }
+
+            payload = {
+                "query": request.keywords,
+                "country": "turkey",
+                "include_domains": ["sigortatahkim.org"],
+                "max_results": request.pageSize,
+                "search_depth": "advanced"
+            }
+
+            if request.page > 1:
+                logger.warning(f"Tavily API doesn't support pagination. Page {request.page} requested.")
+
+            response = await self.http_client.post(
+                self.TAVILY_API_URL,
+                json=payload,
+                headers=headers
+            )
+            response.raise_for_status()
+
+            data = response.json()
+            logger.info(f"Tavily returned {len(data.get('results', []))} results for Sigorta Tahkim")
+
+            decisions = []
+            for result in data.get("results", []):
+                url = result.get("url", "")
+                title = result.get("title", "").strip()
+                content = result.get("content", "")[:500]
+
+                issue_num = self._extract_issue_number(url)
+                doc_id = issue_num if issue_num else url
+
+                decision = SigortaTahkimDecisionSummary(
+                    title=title,
+                    document_id=doc_id,
+                    content=content,
+                    url=url
+                )
+                decisions.append(decision)
+
+            return SigortaTahkimSearchResult(
+                decisions=decisions,
+                total_results=len(data.get("results", [])),
+                page=request.page,
+                pageSize=request.pageSize
+            )
+
+        except httpx.HTTPStatusError as e:
+            logger.error(f"HTTP error searching Sigorta Tahkim decisions: {e}")
+            if e.response.status_code == 401:
+                raise Exception("Tavily API authentication failed. Check API key.")
+            raise Exception(f"Failed to search Sigorta Tahkim decisions: {str(e)}")
+        except Exception as e:
+            logger.error(f"Error searching Sigorta Tahkim decisions: {e}")
+            raise Exception(f"Failed to search Sigorta Tahkim decisions: {str(e)}")
+
+    # Regex pattern to split decisions within a journal issue
+    DECISION_HEADER_PATTERN = re.compile(
+        r'(\d{2}\.\d{2}\.\d{4}\s+Tarih\s+ve\s+K-\d{4}/\d+\s+Sayılı\s+Hakem\s+Kararı)'
+    )
+    # Minimum body length to distinguish real decisions from TOC entries
+    MIN_DECISION_BODY_LENGTH = 1000
+
+    async def _download_and_convert_pdf(self, issue_number: str) -> tuple[str, str]:
+        """
+        Download a journal issue PDF and convert to markdown.
+
+        Returns:
+            Tuple of (markdown_content, pdf_url)
+        """
+        issue_num = int(issue_number)
+        filename = self._get_pdf_filename(issue_num)
+        pdf_url = f"{self.PDF_BASE_URL}{filename}"
+
+        logger.info(f"Downloading Sigorta Tahkim PDF: {pdf_url}")
+
+        response = await self.http_client.get(pdf_url, follow_redirects=True)
+        response.raise_for_status()
+
+        pdf_stream = io.BytesIO(response.content)
+        # markitdown is sync; offload to thread so PDF parsing doesn't block
+        # the event-loop / other in-flight MCP requests.
+        result = await asyncio.to_thread(
+            self.markitdown.convert_stream, pdf_stream, file_extension=".pdf"
+        )
+        return result.text_content.strip(), pdf_url
+
+    def _split_into_decisions(self, markdown_content: str) -> list[tuple[str, str]]:
+        """
+        Split markdown content into individual decisions.
+
+        Returns:
+            List of (header, body) tuples for decisions with substantial content.
+        """
+        parts = self.DECISION_HEADER_PATTERN.split(markdown_content)
+        decisions = []
+        for i in range(1, len(parts) - 1, 2):
+            header = parts[i].strip()
+            body = parts[i + 1].strip() if i + 1 < len(parts) else ""
+            if len(body) >= self.MIN_DECISION_BODY_LENGTH:
+                decisions.append((header, body))
+        return decisions
+
+    async def get_document_markdown(
+        self,
+        issue_number: str,
+        page_number: int = 1
+    ) -> SigortaTahkimDocumentMarkdown:
+        """
+        Retrieve a Sigorta Tahkim journal issue PDF and convert to Markdown.
+
+        Args:
+            issue_number: Journal issue number (e.g., '64')
+            page_number: Page number for paginated content (1-indexed)
+
+        Returns:
+            SigortaTahkimDocumentMarkdown with paginated content
+        """
+        try:
+            markdown_content, pdf_url = await self._download_and_convert_pdf(issue_number)
+
+            total_length = len(markdown_content)
+            total_pages = max(1, math.ceil(total_length / self.DOCUMENT_MARKDOWN_CHUNK_SIZE))
+
+            start_idx = (page_number - 1) * self.DOCUMENT_MARKDOWN_CHUNK_SIZE
+            end_idx = start_idx + self.DOCUMENT_MARKDOWN_CHUNK_SIZE
+            page_content = markdown_content[start_idx:end_idx]
+
+            return SigortaTahkimDocumentMarkdown(
+                document_id=issue_number,
+                markdown_content=page_content,
+                page_number=page_number,
+                total_pages=total_pages,
+                source_url=pdf_url
+            )
+
+        except ValueError:
+            raise Exception(f"Invalid issue number: {issue_number}. Must be a number (e.g., '64').")
+        except httpx.HTTPStatusError as e:
+            logger.error(f"HTTP error fetching Sigorta Tahkim issue {issue_number}: {e}")
+            raise Exception(f"Failed to fetch journal issue {issue_number}: {str(e)}")
+        except Exception as e:
+            logger.error(f"Error processing Sigorta Tahkim issue {issue_number}: {e}")
+            raise Exception(f"Failed to process journal issue {issue_number}: {str(e)}")
+
+    async def search_within_issue(
+        self,
+        issue_number: str,
+        keyword: str,
+        max_results: int = 10
+    ) -> SigortaTahkimSearchWithinResult:
+        """
+        Search for a keyword within a specific journal issue's decisions.
+
+        Downloads the PDF, splits into individual decisions, and returns
+        matching decisions sorted by relevance (match count).
+
+        Args:
+            issue_number: Journal issue number (e.g., '64')
+            keyword: Search keyword or phrase in Turkish
+            max_results: Maximum matching decisions to return
+
+        Returns:
+            SigortaTahkimSearchWithinResult with matching decisions
+        """
+        try:
+            markdown_content, _ = await self._download_and_convert_pdf(issue_number)
+            decisions = self._split_into_decisions(markdown_content)
+
+            logger.info(
+                f"Searching '{keyword}' within issue {issue_number}: "
+                f"{len(decisions)} decisions found"
+            )
+
+            keyword_lower = _turkish_lower(keyword)
+            matches = []
+
+            for header, body in decisions:
+                body_lower = _turkish_lower(body)
+                count = body_lower.count(keyword_lower)
+                if count == 0:
+                    continue
+
+                # Extract excerpt around the first match
+                first_pos = body_lower.find(keyword_lower)
+                excerpt_start = max(0, first_pos - 200)
+                excerpt_end = min(len(body), first_pos + len(keyword) + 200)
+                excerpt = body[excerpt_start:excerpt_end].strip()
+                if excerpt_start > 0:
+                    excerpt = "..." + excerpt
+                if excerpt_end < len(body):
+                    excerpt = excerpt + "..."
+
+                matches.append(SigortaTahkimSearchWithinMatch(
+                    decision_header=header,
+                    relevance_score=count,
+                    excerpt=excerpt,
+                    body_length=len(body)
+                ))
+
+            # Sort by relevance (highest match count first)
+            matches.sort(key=lambda m: m.relevance_score, reverse=True)
+            matches = matches[:max_results]
+
+            return SigortaTahkimSearchWithinResult(
+                issue_number=issue_number,
+                keyword=keyword,
+                total_decisions=len(decisions),
+                matching_decisions=len(matches),
+                matches=matches
+            )
+
+        except ValueError:
+            raise Exception(f"Invalid issue number: {issue_number}. Must be a number (e.g., '64').")
+        except httpx.HTTPStatusError as e:
+            logger.error(f"HTTP error in search_within issue {issue_number}: {e}")
+            raise Exception(f"Failed to fetch journal issue {issue_number}: {str(e)}")
+        except Exception as e:
+            logger.error(f"Error in search_within issue {issue_number}: {e}")
+            raise Exception(f"Failed to search within issue {issue_number}: {str(e)}")

sigorta_tahkim_mcp_module/models.py

@@ -0,0 +1,59 @@
+# sigorta_tahkim_mcp_module/models.py
+
+from pydantic import BaseModel, Field
+from typing import List
+
+
+class SigortaTahkimSearchRequest(BaseModel):
+    """Request model for searching Sigorta Tahkim Komisyonu decisions via Tavily API."""
+    keywords: str = Field(..., description="Search keywords in Turkish")
+    page: int = Field(1, ge=1, description="Page number (1-indexed)")
+    pageSize: int = Field(10, ge=1, le=50, description="Results per page (1-50)")
+
+
+class SigortaTahkimDecisionSummary(BaseModel):
+    """Summary of a Sigorta Tahkim decision from search results."""
+    title: str = Field(..., description="Decision title or journal issue info")
+    document_id: str = Field(..., description="Journal issue number (e.g., '64')")
+    content: str = Field(..., description="Decision summary/excerpt")
+    url: str = Field("", description="Source URL")
+
+
+class SigortaTahkimSearchResult(BaseModel):
+    """Response model for Sigorta Tahkim decision search results."""
+    decisions: List[SigortaTahkimDecisionSummary] = Field(
+        default_factory=list,
+        description="List of matching decisions"
+    )
+    total_results: int = Field(0, description="Total number of results")
+    page: int = Field(1, description="Current page number")
+    pageSize: int = Field(10, description="Results per page")
+
+
+class SigortaTahkimDocumentMarkdown(BaseModel):
+    """Sigorta Tahkim journal issue converted to Markdown format."""
+    document_id: str = Field(..., description="Journal issue number")
+    markdown_content: str = Field("", description="Document content in Markdown")
+    page_number: int = Field(1, description="Current page number")
+    total_pages: int = Field(1, description="Total number of pages")
+    source_url: str = Field("", description="PDF source URL")
+
+
+class SigortaTahkimSearchWithinMatch(BaseModel):
+    """A single matching decision from search within a journal issue."""
+    decision_header: str = Field(..., description="Decision header (date and K-number)")
+    relevance_score: int = Field(0, description="Number of keyword matches")
+    excerpt: str = Field("", description="Matching excerpt with context")
+    body_length: int = Field(0, description="Full decision body length in chars")
+
+
+class SigortaTahkimSearchWithinResult(BaseModel):
+    """Response model for search within a journal issue."""
+    issue_number: str = Field(..., description="Journal issue number searched")
+    keyword: str = Field("", description="Search keyword used")
+    total_decisions: int = Field(0, description="Total decisions in issue")
+    matching_decisions: int = Field(0, description="Number of matching decisions")
+    matches: List[SigortaTahkimSearchWithinMatch] = Field(
+        default_factory=list,
+        description="List of matching decisions sorted by relevance"
+    )

starlette_app.py

@@ -1,159 +0,0 @@
-"""
-Starlette integration example for Yargı MCP Server
-
-This module demonstrates how to integrate the Yargı MCP server
-with a Starlette application, including authentication middleware
-and custom routing.
-
-Usage:
-    uvicorn starlette_app:app --host 0.0.0.0 --port 8000
-"""
-
-import os
-from starlette.applications import Starlette
-from starlette.routing import Mount, Route
-from starlette.requests import Request
-from starlette.responses import JSONResponse, PlainTextResponse, RedirectResponse
-from starlette.middleware import Middleware
-from starlette.middleware.cors import CORSMiddleware
-from starlette.middleware.authentication import AuthenticationMiddleware
-from starlette.authentication import (
-    AuthenticationBackend, AuthCredentials, SimpleUser, AuthenticationError
-)
-
-# Import the main MCP app
-from mcp_server_main import app as mcp_server
-
-# Simple token authentication backend
-class TokenAuthBackend(AuthenticationBackend):
-    async def authenticate(self, request):
-        auth_header = request.headers.get("Authorization")
-        expected_token = os.getenv("API_TOKEN")
-        
-        # Skip auth for health check and public endpoints
-        if request.url.path in ["/health", "/", "/login"]:
-            return None
-            
-        if not expected_token:
-            # No token configured, allow all
-            return AuthCredentials(["authenticated"]), SimpleUser("anonymous")
-            
-        if not auth_header:
-            raise AuthenticationError("Authorization header required")
-            
-        try:
-            scheme, token = auth_header.split()
-            if scheme.lower() != "bearer":
-                raise AuthenticationError("Invalid authentication scheme")
-                
-            if token != expected_token:
-                raise AuthenticationError("Invalid token")
-                
-            return AuthCredentials(["authenticated"]), SimpleUser("user")
-        except ValueError:
-            raise AuthenticationError("Invalid authorization header format")
-
-# Homepage
-async def homepage(request: Request):
-    return JSONResponse({
-        "service": "Yargı MCP Server",
-        "version": "0.1.0",
-        "endpoints": {
-            "mcp": "/mcp-server/mcp/",
-            "api": "/api/",
-            "health": "/health"
-        }
-    })
-
-# API info endpoint
-async def api_info(request: Request):
-    if not request.user.is_authenticated:
-        return JSONResponse({"error": "Authentication required"}, status_code=401)
-        
-    return JSONResponse({
-        "authenticated_as": request.user.display_name,
-        "available_tools": len(mcp_server._tool_manager._tools),
-        "databases": [
-            "Yargıtay", "Danıştay", "Emsal", "Uyuşmazlık",
-            "Anayasa", "KIK", "Rekabet", "Bedesten"
-        ]
-    })
-
-# Health check
-async def health_check(request: Request):
-    return JSONResponse({
-        "status": "healthy",
-        "service": "Yargı MCP Server"
-    })
-
-# Login example (returns token for demo)
-async def login(request: Request):
-    token = os.getenv("API_TOKEN", "demo-token")
-    return JSONResponse({
-        "message": "Use this token in Authorization header",
-        "example": f"Authorization: Bearer {token}",
-        "note": "Set API_TOKEN environment variable to change token"
-    })
-
-# Create MCP ASGI app
-mcp_app = mcp_server.http_app(path='/mcp')
-
-# Configure middleware
-middleware = [
-    Middleware(
-        CORSMiddleware,
-        allow_origins=os.getenv("ALLOWED_ORIGINS", "*").split(","),
-        allow_credentials=True,
-        allow_methods=["*"],
-        allow_headers=["*"],
-    ),
-    Middleware(AuthenticationMiddleware, backend=TokenAuthBackend()),
-]
-
-# Create routes
-routes = [
-    Route("/", homepage),
-    Route("/health", health_check),
-    Route("/login", login),
-    Route("/api/info", api_info),
-    Mount("/mcp-server", app=mcp_app),
-]
-
-# Create Starlette app
-app = Starlette(
-    routes=routes,
-    middleware=middleware,
-    lifespan=mcp_app.lifespan
-)
-
-# Nested mount example
-def create_nested_app():
-    """Example of nested mounting for complex routing structures"""
-    
-    # Create inner app with MCP
-    inner_app = Starlette(
-        routes=[Mount("/services", app=mcp_app)],
-        middleware=middleware
-    )
-    
-    # Create outer app
-    outer_app = Starlette(
-        routes=[
-            Route("/", homepage),
-            Mount("/v1", app=inner_app),
-        ],
-        lifespan=mcp_app.lifespan
-    )
-    
-    # MCP would be available at /v1/services/mcp/
-    return outer_app
-
-# Export both apps
-nested_app = create_nested_app()
-
-if __name__ == "__main__":
-    import uvicorn
-    print("Starting Starlette app with authentication...")
-    print("Set API_TOKEN environment variable to enable authentication")
-    print("Example: API_TOKEN=secret-token python starlette_app.py")
-    uvicorn.run(app, host="0.0.0.0", port=8000)

stripe_webhook.py

@@ -1,25 +0,0 @@
-import os, stripe
-from clerk_backend_api import Clerk                         # Clerk backend SDK
-from fastapi import APIRouter, Request, HTTPException
-
-router = APIRouter()
-stripe.api_key = os.getenv("STRIPE_SECRET")
-clerk         = Clerk(bearer_auth=os.getenv("CLERK_SECRET_KEY"))
-
-@router.post("/stripe/webhook")
-async def stripe_hook(req: Request):
-    payload, sig = await req.body(), req.headers["stripe-signature"]
-    try:
-        event = stripe.Webhook.construct_event(             # Stripe-recommended verify
-            payload, sig, os.getenv("STRIPE_WEBHOOK_SECRET"))
-    except stripe.error.SignatureVerificationError:
-        raise HTTPException(400, "Bad sig")
-
-    if event["type"] == "customer.subscription.updated":
-        item   = event["data"]["object"]["items"]["data"][0]
-        plan   = item["price"]["nickname"]                  # "Pro", "Enterprise"…
-        userID = event["data"]["object"]["metadata"]["clerk_user_id"]
-        clerk.users.update_user_metadata(                   # merge into unsafe_metadata
-            userID, unsafe_metadata={"plan": plan})
-    return {"ok": True}
-    

uyusmazlik_mcp_module/client.py

@@ -1,7 +1,7 @@
 # uyusmazlik_mcp_module/client.py
 
+import asyncio
 import httpx
-import aiohttp 
 from bs4 import BeautifulSoup
 from typing import Dict, Any, List, Optional, Union, Tuple
 import logging
@@ -9,7 +9,7 @@ import html
 import re
 import io
 from markitdown import MarkItDown
-from urllib.parse import urljoin, urlencode # urlencode for aiohttp form data
+from urllib.parse import urljoin
 
 from .models import (
     UyusmazlikSearchRequest,
@@ -56,17 +56,21 @@ class UyusmazlikApiClient:
     # Individual documents are fetched by their full URLs obtained from search results.
 
     def __init__(self, request_timeout: float = 30.0):
-        self.request_timeout = request_timeout # Store timeout for aiohttp and httpx
-        # Headers for aiohttp search. httpx for docs will create its own.
-        self.default_aiohttp_search_headers = {
-            "Accept": "*/*", # Mimicking browser headers provided by user
+        self.request_timeout = request_timeout
+        # Create shared httpx client for all requests
+        self.http_client = httpx.AsyncClient(
+            base_url=self.BASE_URL,
+            headers={
+                "Accept": "*/*",
                 "Accept-Encoding": "gzip, deflate, br, zstd", 
                 "Accept-Language": "tr-TR,tr;q=0.9,en-US;q=0.8,en;q=0.7",
                 "X-Requested-With": "XMLHttpRequest",
                 "Origin": self.BASE_URL,
                 "Referer": self.BASE_URL + "/",
-
-        }
+            },
+            timeout=request_timeout,
+            verify=False
+        )
 
 
     async def search_decisions(
@@ -107,32 +111,36 @@ class UyusmazlikApiClient:
         add_to_form_data("Hepsi", params.hepsi)
         add_to_form_data("Herhangibirisi", params.herhangi_birisi)
         add_to_form_data("NotHepsi", params.not_hepsi)
-        # X-Requested-With is handled by default_aiohttp_search_headers
 
-        search_url = urljoin(self.BASE_URL, self.SEARCH_ENDPOINT)
-        # For aiohttp, data for application/x-www-form-urlencoded should be a dict or str.
-        # Using urlencode for list of tuples.
-        encoded_form_payload = urlencode(form_data_list, encoding='UTF-8') 
+        # Convert form data to dict for httpx
+        form_data_dict = {}
+        for key, value in form_data_list:
+            if key in form_data_dict:
+                # Handle multiple values (like KararSonucuList)
+                if not isinstance(form_data_dict[key], list):
+                    form_data_dict[key] = [form_data_dict[key]]
+                form_data_dict[key].append(value)
+            else:
+                form_data_dict[key] = value
 
-        logger.info(f"UyusmazlikApiClient (aiohttp): Performing search to {search_url} with form_data: {encoded_form_payload}")
-        
-        html_content = ""
-        aiohttp_headers = self.default_aiohttp_search_headers.copy()
-        aiohttp_headers["Content-Type"] = "application/x-www-form-urlencoded; charset=UTF-8"
+        logger.info(f"UyusmazlikApiClient (httpx): Performing search to {self.SEARCH_ENDPOINT} with form_data: {form_data_dict}")
         
         try:
-            # Create a new session for each call for simplicity with aiohttp here
-            async with aiohttp.ClientSession(headers=aiohttp_headers) as session:
-                async with session.post(search_url, data=encoded_form_payload, timeout=self.request_timeout) as response:
-                    response.raise_for_status() # Raises ClientResponseError for 400-599
-                    html_content = await response.text(encoding='utf-8') # Ensure correct encoding
-                    logger.debug("UyusmazlikApiClient (aiohttp): Received HTML response for search.")
+            # Use shared httpx client
+            response = await self.http_client.post(
+                self.SEARCH_ENDPOINT,
+                data=form_data_dict,
+                headers={"Content-Type": "application/x-www-form-urlencoded; charset=UTF-8"}
+            )
+            response.raise_for_status()
+            html_content = response.text
+            logger.debug("UyusmazlikApiClient (httpx): Received HTML response for search.")
         
-        except aiohttp.ClientError as e:
-            logger.error(f"UyusmazlikApiClient (aiohttp): HTTP client error during search: {e}")
+        except httpx.HTTPError as e:
+            logger.error(f"UyusmazlikApiClient (httpx): HTTP client error during search: {e}")
             raise # Re-raise to be handled by the MCP tool
         except Exception as e:
-            logger.error(f"UyusmazlikApiClient (aiohttp): Error processing search request: {e}")
+            logger.error(f"UyusmazlikApiClient (httpx): Error processing search request: {e}")
             raise
 
         # --- HTML Parsing (remains the same as previous version) ---
@@ -217,7 +225,6 @@ class UyusmazlikApiClient:
         try:
             # Using a new httpx.AsyncClient instance for this GET request for simplicity
             async with httpx.AsyncClient(verify=False, timeout=self.request_timeout) as doc_fetch_client:
-
                  get_response = await doc_fetch_client.get(document_url, headers={"Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8"})
             get_response.raise_for_status()
             html_content_from_api = get_response.text
@@ -226,7 +233,7 @@ class UyusmazlikApiClient:
                 logger.warning(f"UyusmazlikApiClient: Received empty or non-string HTML from URL {document_url}.")
                 return UyusmazlikDocumentMarkdown(source_url=document_url, markdown_content=None)
 
-            markdown_content = self._convert_html_to_markdown_uyusmazlik(html_content_from_api)
+            markdown_content = await asyncio.to_thread(self._convert_html_to_markdown_uyusmazlik, html_content_from_api)
             return UyusmazlikDocumentMarkdown(source_url=document_url, markdown_content=markdown_content)
         except httpx.RequestError as e:
             logger.error(f"UyusmazlikApiClient (httpx for docs): HTTP error fetching Uyuşmazlık document from {document_url}: {e}")
@@ -236,5 +243,9 @@ class UyusmazlikApiClient:
             raise
 
     async def close_client_session(self):
-
+        """Close the shared httpx client session."""
+        if hasattr(self, 'http_client') and self.http_client:
+            await self.http_client.aclose()
+            logger.info("UyusmazlikApiClient: HTTP client session closed.")
+        else:
             logger.info("UyusmazlikApiClient: No persistent client session from __init__ to close.")

yargitay_mcp_module/client.py

@@ -1,5 +1,6 @@
 # yargitay_mcp_module/client.py
 
+import asyncio
 import httpx
 from bs4 import BeautifulSoup # Still needed for pre-processing HTML before markitdown
 from typing import Dict, Any, List, Optional
@@ -159,7 +160,7 @@ class YargitayOfficialApiClient:
                 logger.error(f"YargitayOfficialApiClient: 'data' field in API response is not a string or not found (ID: {id}).")
                 raise ValueError("Expected HTML content not found in API response's 'data' field.")
 
-            markdown_content = self._convert_html_to_markdown(html_content_from_api)
+            markdown_content = await asyncio.to_thread(self._convert_html_to_markdown, html_content_from_api)
 
             return YargitayDocumentMarkdown(
                 id=id,