mcp-server-demo

MCP server with OpenAPI validation and analysis tools.

Running the server

Run: uv run python main.py

This starts the server on streamable-http at http://127.0.0.1:8000/mcp. It works locally and behind ngrok or other tunnels with no extra commands or env vars. For Cursor/stdio, use: uv run python main.py stdio.

The validate_openapi tool accepts file paths, inline JSON/YAML, or any public URL. The tool fetches and reads the content over HTTP (no manual download required). Supported URL types include: raw file URLs (e.g. raw.githubusercontent.com/.../ProjectSight-v1.json), Google Drive view or direct download links, and any other public URL that returns JSON or YAML. Both “read the page” and “download” semantics work: the tool GETs the URL and parses the response body.

After validation, the prompt_mcp_developer_context tool provides agent platform ecosystem context, an inferred market segment and intended end users from the API, and two prompts for the agent creator: (1) the job role of the person adding MCP tools for this API, and (2) the task/workflow/job the agent being created will aid in. Pass either the analysis from validate_openapi (when valid) or openapi_input (URL/path/inline) to get these prompts.