Free Tool
A Server Card at /.well-known/mcp.json is how Claude, Cursor, and other agents discover that your product has an MCP server. This tool generates the card and a real, deployable Next.js starter for the server behind it.
Agent discoverability
Claude, Cursor, VS Code, and other MCP clients read /.well-known/mcp.json to find your server and see what it can do.
A real starter, not a stub
The generated route.ts is the exact JSON-RPC/CORS scaffold byfaro.ai's own /mcp server runs on — GET and tools/list work immediately after deploy.
Agents that act, not just read
Where llms.txt tells AI agents about you, MCP lets them do things in your product directly. It's the highest-leverage Agent Infrastructure signal in the AI Readiness Scan.
The Model Context Protocol (MCP) is Anthropic's open standard for connecting AI agents to external tools and data. Where llms.txt and structured data help an AI agent understand your business, MCP lets an agent take real actions in it: look up an order, check availability, book a call, file a ticket. It is the difference between an agent that can describe you and an agent that can transact with you.
A Server Card at yourdomain.com/.well-known/mcp.json is how agents discover that a server exists at all, before they connect. It declares the server's name, description, live endpoint, and — critically — its tools, so an agent evaluating whether to connect can see up front what it will be able to do. Faro's own AI Readiness Scan checks for exactly this file, and for whether the endpoint it points to actually responds.
Publishing a Server Card that points at an endpoint with nothing behind it is worse than publishing nothing: an agent that tries to connect and gets a dead route or a malformed response learns that your integration is broken, not just missing. That is why this tool generates two things, not one — the card, and a real starter server implementing the actual JSON-RPC handshake (initialize, tools/list, tools/call) that MCP clients expect, matching the exact scaffold byfaro.ai runs its own /mcp endpoint on.
The generated tool handlers are deliberately honest: each one is a clearly labeled stub that returns a TODO message rather than fake data. Deploy it as-is and your endpoint responds correctly to discovery calls immediately; fill in the handlers with real logic as you have engineering time, one tool at a time.
The Agent Infrastructure category in the Faro AI Readiness Scan checks for both the Server Card and a live response from the endpoint it declares. A card with no working endpoint scores partial credit and flags the mismatch explicitly; a card backed by a real, responding server is full credit. Pair this with your llms.txt and structured data for the full agent-facing stack: llms.txt for understanding, structured data for verification, MCP for action.