Free OpenAPI → MCP Server Converter
Turn your existing REST API into a server AI agents can call. Paste your OpenAPI or Swagger spec — we generate the MCP tools, a .well-known/mcp.json manifest, and a runnable server. Instantly, no signup. Then let Merlonix keep it healthy.
We map each API operation to an MCP tool, build a .well-known/mcp.json manifest, and generate a runnable server that proxies to your API. Deterministic — nothing is stored.
What does the API-to-MCP converter do?
It reads your OpenAPI 3.x or Swagger 2.0 specification and turns each API operation into a Model Context Protocol (MCP) tool — deriving the tool name from the operationId, the description from the summary, and the input schema from the parameters and request body. It gives you back three things: the MCP tool definitions, a .well-known/mcp.json manifest, and a runnable Node MCP server that proxies calls through to your real API. That server lets an AI agent (Claude, Cursor, or any MCP client) call your existing REST endpoints as tools — without you writing a server by hand.
Why turn a REST API into an MCP server?
AI agents don't call REST APIs directly — they call MCP tools. If your product has a REST API but no MCP server, agents can't use it, and you're invisible to the fastest-growing way software gets integrated. An MCP server is the adapter: it exposes your endpoints as typed, described tools an agent can discover and call. You already documented your API in OpenAPI; this turns that documentation into an agent-callable server.
Do you host it, or do I run it myself?
Today this free tool generates the server code and you run it yourself — download it, set your API base URL and (optionally) an API key, and connect it to your MCP client. Nothing is stored on our side; the conversion is a deterministic transform in your browser session. Hosted conversion — where we run and monitor the MCP server for you — is on our roadmap; run the generator now and tell us if you want it hosted.
My API has hundreds of endpoints — should I convert all of them?
No. An agent's tool selection degrades sharply past roughly 40–50 tools, so exposing 200 endpoints as 200 tools makes the server worse, not better. The converter caps generation and warns you when your API is over the limit. Pick the endpoints agents actually need — the highest-value read and action operations — and expose a focused set. The converter also flags operations missing a description (agents rely on descriptions to choose tools) and write operations (POST/PUT/PATCH/DELETE) you should review before letting an agent call them.
What input formats are supported?
OpenAPI 3.x and Swagger 2.0, as JSON. Most API frameworks expose a JSON document at a path like /openapi.json or /swagger.json — paste that, or give us its URL and we'll fetch it. If your spec is in YAML, export or convert it to JSON first (most OpenAPI tools do this in one command). YAML support is coming.
Is the generated server ready for the July 2026 stateless MCP spec?
Yes. The generated server is built on the official @modelcontextprotocol/sdk rather than a hand-rolled protocol implementation, so as the MCP specification evolves — including the 2026-07-28 revision that makes the protocol stateless — you get the updated transport by upgrading the SDK. Your tool definitions don't change; only the SDK does.
How do I keep the MCP server healthy after I ship it?
Once your MCP server is live, Merlonix monitors it: it runs a real MCP handshake on a schedule and alerts you if it goes down, the handshake starts failing, latency crosses your threshold, or a tool is added, removed, or silently changes schema — the kind of breaking change that takes down every agent calling it. Generate your server here, then run the free MCP Server Health Checker on it and claim the free plan to keep it monitored.