A powerful Node.js server that dynamically converts any OpenAPI/Swagger specification into Model Context Protocol (MCP) tools, making APIs instantly accessible to AI assistants like Claude.

🚀 What This Does

Transform any REST API into AI-friendly tools in seconds:

1. Point to any OpenAPI spec (local file or URL)
2. Get instant MCP tools for schema analysis AND API execution
3. Use with Claude or any MCP-compatible AI assistant

✨ Key Features

• Universal Compatibility: Works with any OpenAPI 2.x or 3.x specification
• Dual Tool Types: Schema analysis tools + Dynamic HTTP execution tools
• Multiple Transports: HTTP server + STDIO for Claude integration
• Auto-Discovery: Automatically generates CRUD tools from your API paths
• Zero Configuration: Just point to a schema and go
• Production Ready: Full error handling, logging, and monitoring

📦 Quick Start

# Install
git clone https://github.com/your-username/openapi-mcp-server.git
cd openapi-mcp-server
npm install

# Configure (pick one)
export OPENAPI_URL="https://petstore.swagger.io/v2/swagger.json"
export OPENAPI_URL="./samples/petstore.yaml"
export OPENAPI_URL="https://your-api.com/swagger.json"

# Start the server
npm start

That's it! Your API is now available as MCP tools.

🛠️ What You Get

Schema Analysis Tools (Always Available)

• list-endpoints - See all API paths and methods
• get-endpoint - Detailed endpoint information
• get-request-body - Request schemas and validation
• get-response-schema - Response formats
• list-components - Schema components and models
• search-schema - Full-text search across the spec
• list-security-schemes - Authentication methods

Dynamic HTTP Execution Tools (Auto-Generated)

For each API endpoint, get semantic tools like:

• search-users - GET /users
• create-user - POST /users
• read-user - GET /users/{id}
• update-user - PUT /users/{id}
• delete-user - DELETE /users/{id}
• execute-request - Generic HTTP execution tool

🎯 Real-World Examples

Petstore API

export OPENAPI_URL="https://petstore.swagger.io/v2/swagger.json"
npm start

Result: 15+ tools including search-pet, create-pet, update-pet, etc.

Your Company API

export OPENAPI_URL="https://your-company.com/api/swagger.json"
export API_BASE_URL="https://your-company.com/api"
export API_TOKEN="your-bearer-token"
npm start

Result: Full API access through semantic tools + Claude integration

Local Development

export OPENAPI_URL="./openapi.yaml"
export API_BASE_URL="http://localhost:3000"
npm start

Result: Local API testing through MCP tools

🔧 Configuration

🤖 Claude Integration

For Desktop Claude

1. Start the MCP client: npm run client
2. Add to your Claude config:

{
  "mcpServers": {
    "openapi": {
      "command": "node",
      "args": ["/path/to/openapi-mcp-server/src/client.js"],
      "env": {
        "OPENAPI_URL": "https://your-api.com/swagger.json"
      }
    }
  }
}

For Web/API Usage

Use the HTTP endpoints directly:

curl -X POST http://localhost:8000/tools/call \
  -H "Content-Type: application/json" \
  -d '{"name": "list-endpoints", "arguments": {}}'

📡 Available Transports

HTTP Server (Web APIs)

• POST /tools/list - List available tools
• POST /tools/call - Execute tools
• GET /sse - Server-Sent Events for real-time
• GET /health - Health monitoring

STDIO (Claude Integration)

• npm run client - Direct MCP protocol communication
• Perfect for Claude Desktop integration
• Full streaming support

🧪 Testing

# Test with included Petstore sample
cp .env.example .env
# Edit .env: OPENAPI_URL=./samples/petstore.yaml
npm start

# Run comprehensive tests
npm test        # HTTP endpoints
npm run test:mcp    # MCP protocol
npm run test:client # Full STDIO client
npm run test:all    # Everything

📁 Project Structure

├── src/
│   ├── server.js           # HTTP server entry point
│   ├── client.js           # STDIO client entry point
│   ├── tools/
│   │   ├── registerTools.js    # Schema analysis tools
│   │   └── registerActions.js  # HTTP execution tools
│   ├── utils/
│   │   ├── schemaLoader.js     # OpenAPI schema loading
│   │   └── toolUtils.js        # Tool management
│   └── routes/             # HTTP route handlers
├── test/                   # Comprehensive test suite
└── samples/                # Example OpenAPI specs

🔍 How It Works

1. Schema Loading: Loads OpenAPI spec from file or URL
2. Tool Generation: Creates MCP tools for both analysis and execution
3. Dynamic Mapping: Maps HTTP operations to semantic tool names
4. Parameter Handling: Automatically handles path/query/body parameters
5. Authentication: Supports Bearer tokens and API keys
6. Transport Agnostic: Works over HTTP or STDIO

🎨 Advanced Usage

Custom Tool Filtering

// Only generate tools for specific paths
const filteredPaths = ['/users', '/orders'];

Authentication Strategies

# Bearer token
export API_TOKEN="your-bearer-token"

# API key in header
export API_KEY="your-api-key"

Multiple APIs

Run multiple instances with different configs:

# API 1
PORT=8001 OPENAPI_URL="api1.yaml" npm start &

# API 2  
PORT=8002 OPENAPI_URL="api2.yaml" npm start &

🚦 Supported APIs

• ✅ OpenAPI 3.x (native support)
• ✅ Swagger 2.x (full compatibility)
• ✅ REST APIs with standard HTTP methods
• ✅ Authentication: Bearer, API Key, OAuth2
• ✅ Parameter Types: Path, Query, Header, Body
• ✅ Content Types: JSON, Form data, URL encoded

📊 Monitoring & Logging

• Built-in health checks at /health
• Structured logging with multiple levels
• Request/response tracking
• Error handling and reporting
• Performance metrics

🔒 Security

• Request validation using Zod schemas
• Bearer token support
• CORS configuration
• Rate limiting (configurable)
• Error sanitization in production

🤝 Contributing

1. Fork the repository
2. Add support for your OpenAPI spec
3. Run the test suite: npm run test:all
4. Submit a pull request

📄 License

MIT License - use this to make any API accessible to AI assistants!

🙏 Attribution

Originally inspired by openapi-introspect by Hannes Junnila. This project expands the concept with dynamic tool generation, HTTP execution, and full MCP protocol support.

Transform any API into AI-accessible tools in minutes, not hours.