toolport-openapi-mcp

Turn any OpenAPI / Swagger spec into an MCP server. One tool per endpoint, zero code.

Point it at a spec URL or file, give it auth, and every operation in the API becomes a tool your AI client can call. Built by the team behind Toolport.

Pointed at Stripe's OpenAPI spec, it generates 587 tools from one URL. (Behind Toolport, your agent sees a handful of meta-tools and searches them, for ~90% fewer tokens.)

npx toolport-openapi-mcp

Quick start

It's configured entirely through environment variables, so it drops into any MCP client. The only required one is the spec:

{
  "mcpServers": {
    "petstore": {
      "command": "npx",
      "args": ["-y", "toolport-openapi-mcp"],
      "env": {
        "OPENAPI_SPEC": "https://petstore3.swagger.io/api/v3/openapi.json"
      }
    }
  }
}

That one URL gives the model 19 tools (getPetById, findPetsByStatus, addPet, ...), each with the right input schema pulled straight from the spec.

Not on npm yet? Until it's published, run from source: clone the repo, npm install && npm run build, then swap the npx line for "command": "node", "args": ["/absolute/path/to/toolport-openapi-mcp/dist/index.js"].

With auth

Most real APIs need a key. Pass the header name and value:

"env": {
  "OPENAPI_SPEC": "https://api.example.com/openapi.json",
  "OPENAPI_AUTH_HEADER": "Authorization",
  "OPENAPI_AUTH_VALUE": "Bearer YOUR_TOKEN"
}

Configuration

Pairs perfectly with Toolport

A real API spec is often 50 to 200+ endpoints. Point this at one and your AI client gets 50 to 200+ tool definitions dumped into the model's context on every request, plus fat JSON responses that bloat it further.

Toolport fixes both halves:

• Lazy discovery collapses those 200 tools into 3 the agent searches on demand (~90% fewer tokens).
• Result-shaping caps oversized API responses so a 10,000-row reply doesn't blow your context.

So this server gives you the whole API, and Toolport makes it cheap to actually use.

How it works

1. Loads and dereferences the spec (via @apidevtools/swagger-parser).
2. Maps each operation to an MCP tool: the operationId becomes the tool name; parameters and the request body become the input schema.
3. On a call, builds the HTTP request (path / query / header params, JSON body, your auth header) and returns the response.

Runs on stdio, no network ports, and your auth token stays in the env you give it.

Contributing

Issues and PRs are welcome. To hack on it: npm install, then npm run build.

License

MIT