MCP 服务器

Noor MCP Code Assistant

Bridges Claude Desktop with a local RAG-powered code agent API, enabling context-aware codebase queries and conversation continuity.

README

🚀 Noor MCP Code Assistant

MCP (Model Context Protocol) Server that bridges Claude Desktop with your RAG-powered Code Agent API.

📋 Overview

This project creates an MCP server that allows Claude Desktop to communicate with your existing RAG Code Agent API running at localhost:8900. When you ask code-related questions in Claude Desktop, it can use this tool to fetch context-aware answers from your codebase.

🎯 Features

✅ Seamless integration with Claude Desktop
✅ Connects to existing RAG API (/api/chat/rag)
✅ Session management for conversation continuity
✅ Comprehensive error handling
✅ Configurable via environment variables
✅ Ready for future agent extensions (SQL, React, Java)

🏗️ Project Structure

noor_mcp_code_assistant_parent/
├── env_container/              # Virtual environment
│   ├── bin/
│   ├── lib/
│   └── ...
└── noor_mcp_code_assistant/    # Project source
    ├── mcp_server.py           # Main MCP server
    ├── config.py               # Configuration management
    ├── requirements.txt        # Dependencies
    ├── .env.example            # Environment template
    ├── .env                    # Your configuration (git-ignored)
    └── README.md               # This file

🚀 Quick Start

1️⃣ Create Virtual Environment

Navigate to the parent folder and create the environment:

cd noor_mcp_code_assistant_parent
python3.11 -m venv env_container

Activate the environment:

# Linux/WSL/macOS
source env_container/bin/activate

# Windows PowerShell
.\env_container\Scripts\Activate

2️⃣ Install Requirements

cd noor_mcp_code_assistant
pip install -r requirements.txt

3️⃣ Configure Environment Variables

Copy the example file and customize:

cp .env.example .env

Edit .env with your settings:

# Required
RAG_API_BASE_URL=http://localhost:8900
RAG_CHAT_ENDPOINT=/api/chat/rag

# Optional
DEFAULT_SESSION_ID=claude-desktop-session
REQUEST_TIMEOUT=120

4️⃣ Configure Claude Desktop

Locate your Claude Desktop configuration file:

Windows:

%APPDATA%\Claude\claude_desktop_config.json

(Usually: C:\Users\<YourUsername>\AppData\Roaming\Claude\claude_desktop_config.json)

Add the MCP server configuration:

{
  "mcpServers": {
    "noor-code-assistant": {
      "command": "wsl",
      "args": [
        "-d", "Ubuntu-20.04",
        "--",
        "/home/<your-username>/noor_mcp_code_assistant_parent/env_container/bin/python",
        "/home/<your-username>/noor_mcp_code_assistant_parent/noor_mcp_code_assistant/mcp_server.py"
      ],
      "env": {
        "PYTHONUNBUFFERED": "1"
      }
    }
  }
}

⚠️ Important: Replace <your-username> with your actual WSL username and adjust paths accordingly.

🔍 Find Your WSL Distribution Name

wsl --list --verbose

Use the exact name shown (e.g., Ubuntu-20.04, Ubuntu, etc.)

5️⃣ Verify Setup

Step 1: Ensure your RAG API is running:

# In WSL terminal
cd /path/to/your/rag-project
source env_container/bin/activate
uvicorn main:app --reload --port 8900

Step 2: Test the RAG API:

curl -X POST http://localhost:8900/api/chat/rag \
  -H "Content-Type: application/json" \
  -d '{"session_id": "test", "message": "hello"}'

Step 3: Restart Claude Desktop completely (check system tray)

Step 4: The noor-code-assistant tool should now appear in Claude Desktop!

💬 Usage Examples

Once configured, you can ask Claude Desktop:

"Use the code assistant to create a C# function that gets employee requests by employee ID"

"Ask the code assistant how pagination is implemented in my project"

"Use code_assistant to show me the DTO structure for Employee entity"

🔮 Adding Future Agents

When you're ready to add more agents (SQL, React, Java):

Step 1: Add endpoints to your RAG API

@app.post("/api/sql/agent")
async def sql_agent(request: SQLAgentRequest):
    # Your SQL agent logic
    pass

Step 2: Update `.env`

SQL_AGENT_ENDPOINT=/api/sql/agent

Step 3: Uncomment tool definitions in `mcp_server.py`

Look for the commented sql_agent tool and handler, then uncomment them.

Step 4: Restart Claude Desktop

🐛 Troubleshooting

Issue	Solution
MCP server not appearing	Check Claude Desktop logs at `%APPDATA%\Claude\logs\`
Connection refused	Ensure RAG API is running on `localhost:8900`
WSL path issues	Use full Linux paths (e.g., `/home/user/...`)
Python not found	Use full path to Python in venv
Timeout errors	Increase `REQUEST_TIMEOUT` in `.env`

📋 View Logs

Claude Desktop Logs:

C:\Users\<YourUsername>\AppData\Roaming\Claude\logs\

Test MCP Server Standalone:

cd noor_mcp_code_assistant
source ../env_container/bin/activate
python mcp_server.py

📊 Configuration Reference

Variable	Default	Description
`RAG_API_BASE_URL`	`http://localhost:8900`	Base URL of your RAG API
`RAG_CHAT_ENDPOINT`	`/api/chat/rag`	Chat endpoint path
`DEFAULT_SESSION_ID`	`claude-desktop-session`	Default session for continuity
`REQUEST_TIMEOUT`	`120`	API timeout in seconds