mcpify

Turn any REST API into an MCP server in one command. Point mcpify at an OpenAPI spec and every endpoint becomes a tool your AI agent (Claude, Cursor, Windsurf, …) can call — no glue code, no per-endpoint wrappers.

mcpify https://api.example.com/openapi.json

That's it. Every operation in the spec is now a live MCP tool.

---

Why

MCP is how AI agents call real tools. But wiring an existing API into MCP means hand-writing a tool wrapper for every endpoint — tedious and stale the moment the API changes. Almost every API already publishes an OpenAPI spec. mcpify reads that spec and generates the whole MCP server at runtime, so:

• Zero per-endpoint code — N endpoints → N tools, automatically.
• Always in sync — regenerate from the spec; no wrappers to maintain.
• Auth-aware — pass an API key / bearer token once, applied to every call.

Install

pipx install mcpify-cli      # or: uvx mcpify-cli ...

Use

Preview the tools a spec produces (no server):

mcpify --list examples/petstore-mini.yaml

Run it as an MCP server (stdio):

mcpify https://petstore3.swagger.io/api/v3/openapi.json
mcpify ./openapi.yaml --base-url https://staging.internal.api
mcpify ./openapi.yaml --auth "Authorization: Bearer $TOKEN"   # or set MCPIFY_AUTH

Plug it into Claude

Add to your MCP client config (Claude Desktop / Claude Code mcp.json):

{
  "mcpServers": {
    "petstore": {
      "command": "mcpify",
      "args": ["https://petstore3.swagger.io/api/v3/openapi.json"]
    }
  }
}

Restart the client and ask: "list the available pets" — Claude calls the API directly.

How it works

1. Load the OpenAPI 3.x document (URL or file, JSON or YAML).
2. Walk every path/method into an Operation with a generated JSON input schema.
3. Serve them over MCP; each tool call is mapped to a live HTTP request (path, query, header params and JSON bodies all handled) and the response is returned to the agent.

Supported

• OpenAPI 3.x, JSON or YAML, from a URL or local file
• GET / POST / PUT / PATCH / DELETE
• Path, query, and header parameters; JSON request bodies
• A single global auth header (API key or bearer token)

Roadmap

• Per-operation filtering (--only, --tag) to expose a subset of a large API
• $ref resolution for fully-expanded body schemas
• Swagger 2.0 specs
• SSE / HTTP transport in addition to stdio

PRs welcome.

License

MIT