MCP Router

# Cursor IDE Integration Guide This guide shows how to integrate the MCP Router with Cursor IDE so it automatically selects the best model for each query. ## Overview The integration works by: 1. **Query comes in** → Cursor receives your prompt 2. **Router analyzes** → Determines best model based on query characteristics 3. **Model selected** → Router picks optimal model (GPT-4o, Claude, etc.) 4. **Cursor executes** → Uses the selected model for your prompt ## Setup ### 1. Install Dependencies ```bash cd mcp-router pip install -r requirements.txt pip install mcp # MCP SDK for Cursor integration ``` ### 2. Configure Cursor Add the MCP server to Cursor's configuration: **On macOS:** ```bash # Edit Cursor's MCP settings open ~/Library/Application\ Support/Cursor/User/globalStorage/mcp.json ``` **On Windows:** ``` %APPDATA%\Cursor\User\globalStorage\mcp.json ``` **On Linux:** ``` ~/.config/Cursor/User/globalStorage/mcp.json ``` ### 3. Add MCP Server Configuration Add this to your Cursor MCP configuration file: ```json { "version": "1.0", "mcpServers": { "mcp-router": { "command": "python3", "args": [ "/absolute/path/to/mcp-router/src/mcp_server.py" ], "env": {} } } } ``` **Important:** Replace `/absolute/path/to/mcp-router/src/mcp_server.py` with the actual absolute path to the file. ### 4. Alternative: Use Environment Variables You can also set API keys in your shell environment or `.env` file: ```bash export OPENAI_API_KEY="your_key_here" export ANTHROPIC_API_KEY="your_key_here" ``` Then reference the `.env` file in the MCP server script. ## Usage in Cursor Once configured, Cursor will automatically: 1. **Analyze your queries** using the router 2. **Select the best model** based on: - Query complexity - Task type (reasoning, code, etc.) - Your preferences (cost, speed, quality) 3. **Route to the selected model** automatically ### Example Workflow 1. You type: "Explain how neural networks work" - Router analyzes → Reasoning task, high complexity - Router selects → Claude 3.5 Sonnet (best reasoning) - Cursor uses Claude 3.5 Sonnet for response 2. You type: "Write a Python function to sort a list" - Router analyzes → Code generation, low complexity - Router selects → GPT-4o-mini (fast, cost-effective) - Cursor uses GPT-4o-mini for response 3. You type: "Design a distributed system architecture" - Router analyzes → Complex reasoning task - Router selects → GPT-4o or Claude 3.5 Sonnet (high quality) - Cursor uses selected model for response ## Manual Model Selection You can also manually request routing by using MCP tools in Cursor: - `route_query`: Get routing recommendation without executing - `execute_routed_query`: Route and execute query - `list_models`: See all available models - `get_routing_stats`: View routing statistics - `analyze_query`: Analyze query characteristics ## Troubleshooting ### MCP Server Not Starting 1. Check Python path: Make sure `python3` is in your PATH 2. Check file path: Use absolute path to `mcp_server.py` 3. Check permissions: Ensure the script is executable ### Models Not Available 1. Check API keys: Ensure keys are set in environment or config 2. Check model registration: Verify models are registered in router 3. Check logs: Look for error messages in Cursor's MCP logs ### Routing Not Working 1. Verify MCP connection: Check Cursor's MCP status 2. Test router directly: Run `python main.py route "test query"` 3. Check configuration: Verify MCP server config is correct ## Advanced Configuration ### Custom Routing Strategies You can customize routing by modifying the router configuration: ```python # In mcp_server.py, customize the router router = MCPRouter() # Add custom models router.register_model(ModelCapabilities(...)) # Modify routing logic in router.py ``` ### Environment-Specific Models Set different models for different environments: ```json { "mcpServers": { "mcp-router": { "command": "python3", "args": ["/path/to/mcp_server.py"], "env": { "OPENAI_API_KEY": "${OPENAI_API_KEY}", "ANTHROPIC_API_KEY": "${ANTHROPIC_API_KEY}", "ROUTING_STRATEGY": "balanced" } } } } ``` ## Testing the Integration 1. **Test MCP Server Directly:** ```bash python3 src/mcp_server.py ``` 2. **Test Router:** ```bash python3 main.py route "test query" ``` 3. **Test in Cursor:** - Open Cursor - Check MCP status (should show "mcp-router" connected) - Try a query and observe model selection ## Benefits - **Automatic Optimization**: Always uses the best model for each task - **Cost Efficiency**: Routes to cheaper models when appropriate - **Speed Optimization**: Uses faster models for simple tasks - **Quality Focus**: Uses best models for complex reasoning - **Transparency**: See which model was selected and why ## Next Steps - Customize model configurations for your use case - Add more models to the registry - Implement custom routing strategies - Monitor routing statistics to optimize performance