It is time to re-implement dinnercaster, yet again. This time we will expose a dinnercaster api as a MCP server to connect to claude.ai desktop/web. The MCP inspector tool can connect, but connecting from claude.ai is still not solved.
Where we landed: https://github.com/redfive410/dinnercaster3/pull/1/files
Helpful tool: https://github.com/modelcontextprotocol/inspector
I've successfully updated your repository to expose a route for Model Context Protocol. Here's what I've implemented:
Dependencies Added:
- mcp==1.0.0 - Core MCP library
- pydantic==2.5.0 - For data validation (required by MCP)
MCP Server Features:
- Route: /mcp - Main MCP server endpoint
- Info Route: /mcp-info - Provides information about available MCP tools
- Two Sample Tools:
- echo - Echoes back input text
- get_info - Returns information about the service
Testing the MCP Server:
1. Start the server: uvicorn main:app --reload
2. Visit http://localhost:8000/mcp-info to see available tools
3. The MCP server is mounted at http://localhost:8000/mcp
The implementation provides a foundation for MCP server functionality that can be extended with additional tools as needed.