Work with MCP server

Use the MCP server when you need to implement or modify Model Context Protocol tool handlers for Attune AI workflows.

Prerequisites

• Access to the project source code
• Familiarity with the files under src/attune/mcp/
• Understanding of MCP (Model Context Protocol) concepts

Steps

1. Identify the server component to modify

   Determine which part of the MCP server handles your use case:

   • Prompts: Use src/attune/mcp/prompts.py for prompt templates
   • Tools: Use src/attune/mcp/tool_schemas.py for tool definitions
   • Server core: Use src/attune/mcp/server.py for server lifecycle
   • Memory handlers: Use src/attune/mcp/memory_handlers.py for memory operations
   • Rate limiting: Use src/attune/mcp/rate_limiter.py for call throttling

2. Review the existing implementation

   Open the relevant file and examine:

   • The function signature and docstring
   • Input parameters and return types
   • Error handling patterns
   • Integration with other components

3. Create your MCP server instance

   If working with a new server instance:

   from attune.mcp.server import create_server
   server = create_server()
   

4. Modify the appropriate handler

   Make your changes following the established patterns:

   • Tool functions return dict[str, Any] with structured responses
   • Use the rate limiter for external API calls
   • Follow the naming convention for new tools
   • Include proper error handling with meaningful messages

5. Test your changes

   Run the MCP server tests to verify your implementation:

   pytest -k "mcp" --verbose
   

Verify success

Your implementation is working when:

• The MCP server starts without errors using main()
• Your tool appears in the tool list from get_tool_list()
• Tool calls return the expected response format
• Rate limiting prevents excessive API usage
• All existing tests continue to pass