MCPify

Every AI agent framework exposes tools differently, and wiring a given agent into an MCP client (Claude Desktop, Cursor, VS Code) means reading its docs and hand-writing a config. There was no quick way to point at an arbitrary agent URL and get a working MCP setup.

Built a tool that probes an agent's common endpoints, infers which framework it's built on and with what confidence, then generates the exact JSON config each MCP client expects — and made the tool itself consumable as an MCP server, so the workflow of adding an agent to your client can happen entirely through tool calls.

1. 01FastAPI application (main.py) as the entry point, with fastapi-mcp mounting an /mcp endpoint alongside the REST API
2. 02analyzer.py concurrently probes an agent's common routes and applies framework-detection heuristics
3. 03generator.py turns a detected agent into ready-to-use config JSON for each supported MCP client
4. 04mcp_handler.py defines the three MCP tools (analyze_agent, generate_mcp_config, get_integration_guide) and the REST router backing them
5. 05APScheduler runs a background keep-alive job pinging /health to avoid free-tier cold starts

• ▸Designed the endpoint-probing heuristics to run concurrently across candidate routes for fast analysis
• ▸Built framework-detection scoring so results report a confidence level instead of a binary guess
• ▸Wired fastapi-mcp so the same FastAPI app serves both a normal REST API and an MCP-compliant tool interface
• ▸Wrote a Render deployment blueprint (render.yaml) and a keep-alive worker for free-tier hosting

• ▸Detecting a framework reliably from the outside, using only HTTP responses and heuristics, without access to source code
• ▸Making the tool dual-purpose — usable as a normal web API and as a spec-compliant MCP server — without duplicating logic

A working analyzer and config generator for the MCP ecosystem that is itself a live example of an MCP-compliant server, actively being developed.