mattt/emcee

MCP generator for OpenAPIs 🫳🎤💥

What it solves

emcee solves the problem of manually creating Model Context Protocol (MCP) servers for every web application. It allows users to instantly turn any web application with an OpenAPI specification into an MCP server, enabling AI models (like Claude Desktop) to interact with external tools and data services without needing a custom-built server implementation.

How it works

emcee acts as a bridge between an AI client and a web API. It takes an OpenAPI specification (via URL or local file) and implements the MCP stdio transport using JSON-RPC 2.0. It maps the API endpoints defined in the OpenAPI spec to MCP tools that the AI model can call. It also supports various authentication methods (Bearer tokens, Basic auth, and raw headers) and integrates with 1Password for secure secret management.

Who it’s for

  • Developers building web apps: Those who want to provide AI access to their services without building a dedicated MCP server or client library.
  • AI Power Users: Users of MCP-compatible clients (like Claude Desktop) who want to connect to any API that provides an OpenAPI spec.

Highlights

  • Instant MCP Server: Converts OpenAPI specs directly into functional AI tools.
  • Secure Authentication: Supports Bearer, Basic, and raw auth, with native 1Password CLI integration for secrets.
  • OpenAPI 3.2 Support: Includes support for the HTTP QUERY method (RFC 10008).
  • Flexible Transformation: Works with Unix utilities like jq or yq to filter or modify which API endpoints are exposed as tools.
  • Multi-platform: Available via Homebrew, Docker, or as a pre-built binary for Linux and macOS.