apitomcp 0.1.0

Convert any API documentation into an MCP server

api-to-mcp

Convert any API documentation into an MCP server for Cursor in minutes.

Point it at API docs, and it scrapes, generates an OpenAPI spec, and creates a fully functional MCP server you can chat with.

Quickstart

# Install
pip install apitomcp

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

# Generate an MCP server from any API docs
apitomcp generate

# Install to Cursor
apitomcp install

Restart Cursor 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
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 (used by Cursor, not manually)

Example

$ apitomcp generate

# Enter: https://developer.spotify.com/documentation/web-api
# It scrapes 150+ pages, finds 99 API operations
# Generates OpenAPI spec in parallel
# Detects OAuth2 client credentials auth
# Prompts for your Spotify client ID and secret

$ apitomcp install

# Adds the server to Cursor's MCP config
# Restart Cursor, then ask: "Get me Taylor Swift's top tracks"

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 - Access to Claude, GPT, Gemini models
• Anthropic - Claude Sonnet, Haiku, Opus
• OpenAI - GPT-5.2, GPT-5.2 Mini
• Gemini - Gemini 3 Pro, Gemini 2.5 Flash

Requirements

• Python 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

License

MIT