API To MCP

Convert any API documentation into an MCP server in minutes.

Point it at API docs, and it scrapes, generates an OpenAPI spec, creates a fully functional MCP server you can chat with, and installs it to Cursor or Claude Desktop.

Installation

pip install apitomcp

Or with uv:

uv pip install apitomcp

Quickstart

# Install
pip install apitomcp

# First-time setup (configure your LLM provider)
apitomcp init

# Generate an MCP server from any API docs
apitomcp generate

Restart Cursor/ Claude Desktop and start chatting with your API.

What It Does

1. Scrapes API documentation (handles multi-page docs, finds all endpoints)
2. Generates OpenAPI 3.1 specs using LLMs (parallel processing for speed)
3. Detects authentication requirements (OAuth2, Bearer tokens, API keys)
4. Creates MCP servers that Cursor can use as tools
5. Handles OAuth2 token refresh automatically

Features

• Multi-page scraping - Crawls linked pages to find all API endpoints
• Parallel LLM processing - Generates specs for many endpoints simultaneously
• Smart auth detection - Analyzes docs to detect OAuth2, Bearer, or API key auth
• OAuth2 token refresh - Automatically refreshes expired tokens
• Interactive CLI - Clean prompts for configuration
• Multiple LLM providers - OpenRouter, Anthropic, OpenAI, Gemini

Commands

Command	Description
apitomcp init	First-time setup - configure LLM provider and API key
apitomcp generate	Generate an MCP server from API documentation
apitomcp list	Show all generated servers
apitomcp install	Install servers to Cursor or Claude Desktop
apitomcp delete	Remove a generated server
apitomcp auth	Update LLM settings
apitomcp output	Export server files to current directory
apitomcp run <name>	Run a server (typically used by Cursor or Claude Desktop, not manually)

Example

$ apitomcp generate

# Enter: https://developer.themoviedb.org/docs/
# Enter your API Read Access Token from https://www.themoviedb.org/settings/api

Then chat with your API in Cursor:

How It Works

API Docs URL
     │
     ▼
┌─────────────┐
│   Scraper   │  BeautifulSoup + MarkItDown
│  (multi-page)│  Extracts endpoints & auth info
└─────────────┘
     │
     ▼
┌─────────────┐
│  Generator  │  LLM generates OpenAPI specs
│  (parallel) │  for each endpoint
└─────────────┘
     │
     ▼
┌─────────────┐
│  Validator  │  Validates against OpenAPI 3.1
│             │  Auto-retries on errors
└─────────────┘
     │
     ▼
┌─────────────┐
│   Runner    │  FastMCP creates tools from spec
│             │  Handles auth & token refresh
└─────────────┘

Supported LLM Providers

• OpenRouter
• Anthropic
• OpenAI
• Gemini

Requirements

• Python 3.10-3.13
• API key for one of the supported LLM providers

Configuration

Config is stored in ~/.apitomcp/:

• config.json - LLM provider settings
• servers/<name>/ - Generated server files

Limitations

This is a work in progress. The LLM-based operation extraction and auth detection aren't perfect—some APIs may need manual tweaking of the generated spec. If you run into issues or have improvements, Issues/ PRs are welcome!

Contributing

Contributions are welcome! If you find a bug or have a feature request, please open an issue. Pull requests are also appreciated—I'll review them as time allows.

Author

Built by @andrew_masek_

License

MIT