OpenAPI to Model Context Protocol (MCP)

The OpenAPI-MCP proxy translates OpenAPI specs into MCP tools, enabling AI agents to access external APIs without custom wrappers!

Bridge the gap between AI agents and external APIs

The OpenAPI to Model Context Protocol (MCP) proxy server bridges the gap between AI agents and external APIs by dynamically translating OpenAPI specifications into standardized MCP tools, resources, and prompts. This simplifies integration by eliminating the need for custom API wrappers.

Built with FastMCP following official MCP patterns and best practices, the server provides:

• ✅ Official FastMCP Integration - Uses the latest FastMCP framework for optimal performance

• ✅ Proper MCP Transport - Supports stdio, SSE, and streamable HTTP transports

• ✅ Modular Architecture - Clean separation of concerns with dependency injection

• ✅ Production Ready - Robust error handling, comprehensive logging, and type safety

• Repository: https://github.com/gujord/OpenAPI-MCP

---

If you find it useful, please give it a ⭐ on GitHub!

---

Key Features

Core Functionality

• FastMCP Transport: Optimized for stdio, working out-of-the-box with popular LLM orchestrators.
• OpenAPI Integration: Parses and registers OpenAPI operations as callable tools.
• Resource Registration: Automatically converts OpenAPI component schemas into resource objects with defined URIs.
• Prompt Generation: Generates contextual prompts based on API operations to guide LLMs in using the API.
• Dual Authentication: Supports both OAuth2 Client Credentials flow and username/password authentication with automatic token caching.
• MCP HTTP Transport: Official MCP-compliant HTTP streaming transport with JSON-RPC 2.0 over SSE.
• Server-Sent Events (SSE): Legacy streaming support (deprecated - use MCP HTTP transport).
• JSON-RPC 2.0 Support: Fully compliant request/response structure.

Advanced Features

• Modular Architecture: Clean separation of concerns with dedicated modules for authentication, request handling, and tool generation.
• Robust Error Handling: Comprehensive exception hierarchy with proper JSON-RPC error codes and structured error responses.
• Auto Metadata: Derives tool names, summaries, and schemas from the OpenAPI specification.
• Sanitized Tool Names: Ensures compatibility with MCP name constraints.
• Flexible Parameter Parsing: Supports query strings, JSON, and comma-separated formats with intelligent type conversion.
• Enhanced Parameter Handling: Automatically converts parameters to correct data types with validation.
• Extended Tool Metadata: Includes detailed parameter information, response schemas, and API categorization.
• CRUD Operation Detection: Automatically identifies and generates example prompts for Create, Read, Update, Delete operations.
• MCP-Compliant Streaming: Official MCP HTTP transport for real-time streaming with proper session management.

Developer Experience

• Configuration Management: Centralized environment variable handling with validation and defaults.
• Comprehensive Logging: Structured logging with appropriate levels for debugging and monitoring.
• Type Safety: Full type hints and validation throughout the codebase.
• Extensible Design: Factory patterns and dependency injection for easy customization and testing.

🚀 Quick Start

Installation

Option 1: Using uvx (Recommended)

# Run directly without installation
uvx openapi-mcp-proxy

# Or with environment variables
OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
uvx openapi-mcp-proxy

Option 2: Using pip

pip install openapi-mcp-proxy

# Then run
OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
openapi-mcp

Option 3: From source

git clone https://github.com/gujord/OpenAPI-MCP.git
cd OpenAPI-MCP
python3.12 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -e .

🎯 Simple Usage

Quick Test (Norwegian Weather API)

# Using uvx
OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
uvx openapi-mcp

# Or using installed package
OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
openapi-mcp

HTTP Transport (Recommended for Claude Desktop)

# Start weather API with HTTP transport
OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
MCP_HTTP_ENABLED="true" \
MCP_HTTP_PORT="8001" \
openapi-mcp

🔗 Claude Desktop Setup

1. Copy the provided configuration:

cp claude_desktop_config.json ~/Library/Application\ Support/Claude/claude_desktop_config.json

2. Start the weather server:

OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
MCP_HTTP_ENABLED="true" \
MCP_HTTP_PORT="8001" \
openapi-mcp

3. Test in Claude Desktop:

• Ask: "What's the weather in Oslo tomorrow?"
• Claude will use the weather_get__compact tool automatically!

🌐 Multiple API Servers

Run multiple OpenAPI services simultaneously:

# Terminal 1: Weather API
source venv/bin/activate && \
OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
MCP_HTTP_ENABLED="true" \
MCP_HTTP_PORT="8001" \
python src/openapi_mcp/fastmcp_server.py

# Terminal 2: Petstore API  
source venv/bin/activate && \
OPENAPI_URL="https://petstore3.swagger.io/api/v3/openapi.json" \
SERVER_NAME="petstore" \
MCP_HTTP_ENABLED="true" \
MCP_HTTP_PORT="8002" \
python src/openapi_mcp/fastmcp_server.py

🐳 Docker Deployment

Quick start with Docker:

# Start all services (weather + petstore)
./docker-start.sh

# Or manually
docker-compose up --build -d

This automatically runs:

• Weather API on port 8001
• Petstore API on port 8002

⚙️ Advanced Configuration

Claude Desktop / Cursor / Windsurf

HTTP Transport (Recommended):

Use the provided configuration file:

cp claude_desktop_config.json ~/Library/Application\ Support/Claude/claude_desktop_config.json

Or create manually:

{
  "mcpServers": {
    "weather": {
      "command": "npx",
      "args": ["mcp-remote", "http://127.0.0.1:8001/sse"]
    },
    "petstore": {
      "command": "npx", 
      "args": ["mcp-remote", "http://127.0.0.1:8002/sse"]
    }
  }
}

Stdio Transport (Alternative):

{
  "mcpServers": {
    "weather": {
      "command": "/full/path/to/OpenAPI-MCP/venv/bin/python",
      "args": ["/full/path/to/OpenAPI-MCP/src/openapi_mcp/fastmcp_server.py"],
      "env": {
        "SERVER_NAME": "weather",
        "OPENAPI_URL": "https://api.met.no/weatherapi/locationforecast/2.0/swagger"
      },
      "transport": "stdio"
    }
  }
}

Note: Replace /full/path/to/OpenAPI-MCP with your actual installation path.

With Local Spec and Custom Headers

{
  "mcpServers": {
    "local_api": {
      "command": "/full/path/to/OpenAPI-MCP/venv/bin/python",
      "args": ["/full/path/to/OpenAPI-MCP/src/openapi_mcp/fastmcp_server.py"],
      "env": {
        "SERVER_NAME": "local_api",
        "OPENAPI_URL": "./specs/my-api.yaml",
        "MCP_AUTH_HEADERS": "{\"X-API-Key\": \"your-key-here\"}"
      },
      "transport": "stdio"
    }
  }
}

With Username/Password Authentication

{
  "mcpServers": {
    "secure_api": {
      "command": "full_path_to_openapi_mcp/venv/bin/python",
      "args": ["full_path_to_openapi_mcp/src/openapi_mcp/fastmcp_server.py"],
      "env": {
        "SERVER_NAME": "secure_api",
        "OPENAPI_URL": "https://api.example.com/openapi.json",
        "API_USERNAME": "your_username",
        "API_PASSWORD": "your_password"
      },
      "transport": "stdio"
    }
  }
}

With OAuth2 Authentication

{
  "mcpServers": {
    "oauth_api": {
      "command": "full_path_to_openapi_mcp/venv/bin/python",
      "args": ["full_path_to_openapi_mcp/src/openapi_mcp/fastmcp_server.py"],
      "env": {
        "SERVER_NAME": "oauth_api",
        "OPENAPI_URL": "https://api.example.com/openapi.json",
        "OAUTH_CLIENT_ID": "your_client_id",
        "OAUTH_CLIENT_SECRET": "your_client_secret",
        "OAUTH_TOKEN_URL": "https://api.example.com/oauth/token"
      },
      "transport": "stdio"
    }
  }
}

Multiple API Servers with MCP HTTP Transport

Configure multiple OpenAPI services to run simultaneously:

{
  "mcpServers": {
    "weather": {
      "command": "npx",
      "args": [
        "mcp-remote", 
        "http://127.0.0.1:8001/sse"
      ]
    },
    "petstore": {
      "command": "npx", 
      "args": [
        "mcp-remote",
        "http://127.0.0.1:8002/sse" 
      ]
    }
  }
}

This configuration gives Claude access to both weather data AND petstore API tools simultaneously, with clear tool naming like weather_get__compact and petstore_addPet.

Single API Server with MCP HTTP Transport

For a single API service:

Standard SSE Configuration:

{
  "mcpServers": {
    "openapi_service": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://127.0.0.1:8001/sse"
      ]
    }
  }
}

Streamable HTTP Configuration:

{
  "mcpServers": {
    "openapi_service": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://127.0.0.1:8001/mcp"
      ]
    }
  }
}

With Debugging (for development):

{
  "mcpServers": {
    "openapi_service": {
      "command": "npx",
      "args": [
        "mcp-remote", 
        "http://127.0.0.1:8001/sse",
        "--debug"
      ]
    }
  }
}

With Custom Transport Strategy:

{
  "mcpServers": {
    "openapi_service": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "http://127.0.0.1:8001/mcp", 
        "--transport",
        "streamable-http"
      ]
    }
  }
}

With Legacy SSE Streaming (Deprecated)

{
  "mcpServers": {
    "streaming_api": {
      "command": "full_path_to_openapi_mcp/venv/bin/python",
      "args": ["full_path_to_openapi_mcp/src/openapi_mcp/fastmcp_server.py"],
      "env": {
        "SERVER_NAME": "streaming_api",
        "OPENAPI_URL": "https://api.example.com/openapi.json",
        "SSE_ENABLED": "true",
        "SSE_HOST": "127.0.0.1",
        "SSE_PORT": "8001"
      },
      "transport": "stdio"
    }
  }
}

Apply this configuration to the following files:

• Cursor: ~/.cursor/mcp.json
• Windsurf: ~/.codeium/windsurf/mcp_config.json
• Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json

Replace full_path_to_openapi_mcp with your actual installation path.

Quick Setup for Multiple APIs

Copy the provided example configuration:

cp claude_desktop_config.json ~/Library/Application\ Support/Claude/claude_desktop_config.json

Start both services:

# Terminal 1
source venv/bin/activate && \
OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
MCP_HTTP_ENABLED="true" \
MCP_HTTP_PORT="8001" \
python src/openapi_mcp/fastmcp_server.py

# Terminal 2  
source venv/bin/activate && \
OPENAPI_URL="https://petstore3.swagger.io/api/v3/openapi.json" \
SERVER_NAME="petstore" \
MCP_HTTP_ENABLED="true" \
MCP_HTTP_PORT="8002" \
python src/openapi_mcp/fastmcp_server.py

Result: Claude gets access to both weather and petstore APIs with prefixed tool names.

Environment Configuration

Core Configuration

Variable	Description	Required	Default
OPENAPI_URL	URL or local file path to OpenAPI specification	Yes	-
SERVER_NAME	MCP server name	No	openapi_proxy_server

OAuth2 Authentication

Variable	Description	Required	Default
OAUTH_CLIENT_ID	OAuth client ID	No	-
OAUTH_CLIENT_SECRET	OAuth client secret	No	-
OAUTH_TOKEN_URL	OAuth token endpoint URL	No	-
OAUTH_SCOPE	OAuth scope	No	api

Username/Password Authentication

Variable	Description	Required	Default
API_USERNAME	API username for authentication	No	-
API_PASSWORD	API password for authentication	No	-
API_LOGIN_ENDPOINT	Login endpoint URL	No	Auto-detected

Custom Authentication Headers

Variable	Description	Required	Default
MCP_AUTH_HEADERS	Custom authentication headers (JSON or key=value format)	No	-

MCP HTTP Transport (Recommended)

Variable	Description	Required	Default
MCP_HTTP_ENABLED	Enable MCP HTTP transport	No	false
MCP_HTTP_HOST	MCP HTTP server host	No	127.0.0.1
MCP_HTTP_PORT	MCP HTTP server port	No	8000
MCP_CORS_ORIGINS	CORS origins (comma-separated)	No	*
MCP_MESSAGE_SIZE_LIMIT	Message size limit	No	4mb
MCP_BATCH_TIMEOUT	Batch timeout in seconds	No	30
MCP_SESSION_TIMEOUT	Session timeout in seconds	No	3600

Legacy SSE Support (Deprecated)

Variable	Description	Required	Default
SSE_ENABLED	Enable SSE streaming support	No	false
SSE_HOST	SSE server host	No	127.0.0.1
SSE_PORT	SSE server port	No	8000

📁 Loading Local OpenAPI Specifications

You can now load OpenAPI specs from your local filesystem instead of requiring remote URLs:

JSON Format

source venv/bin/activate
OPENAPI_URL="./specs/my-api.json" \
SERVER_NAME="local_api" \
python src/openapi_mcp/fastmcp_server.py

YAML Format

source venv/bin/activate
OPENAPI_URL="../shared/api.yaml" \
SERVER_NAME="local_api" \
python src/openapi_mcp/fastmcp_server.py

Absolute Path

source venv/bin/activate
OPENAPI_URL="/Users/myuser/projects/api-spec.json" \
SERVER_NAME="local_api" \
python src/openapi_mcp/fastmcp_server.py

Supported Formats

• JSON files: .json extension
• YAML files: .yaml or .yml extension
• Relative paths: ./path/to/spec.yaml, ../spec.json
• Absolute paths: /full/path/to/spec.yaml

🔑 API Key and Custom Headers

Support for APIs requiring custom headers (API keys, tokens, etc.):

Using JSON Format

source venv/bin/activate
MCP_AUTH_HEADERS='{"X-API-Key": "your-api-key", "X-Client-ID": "client123"}' \
OPENAPI_URL="https://api.example.com/openapi.json" \
SERVER_NAME="custom_api" \
python src/openapi_mcp/fastmcp_server.py

Using Simple Format

source venv/bin/activate
MCP_AUTH_HEADERS='X-API-Key=your-api-key,X-Client-ID=client123' \
OPENAPI_URL="https://api.example.com/openapi.json" \
SERVER_NAME="custom_api" \
python src/openapi_mcp/fastmcp_server.py

Common API Key Patterns

• RapidAPI: MCP_AUTH_HEADERS='{"X-RapidAPI-Key": "your-key"}'
• Custom Bearer: MCP_AUTH_HEADERS='{"Authorization": "Bearer custom-token"}'
• Multiple Headers: MCP_AUTH_HEADERS='{"X-API-Key": "key", "X-API-Secret": "secret"}'

Local Spec + Custom Headers

Combine both features for development:

source venv/bin/activate
OPENAPI_URL="./test/fixtures/api.json" \
MCP_AUTH_HEADERS='{"X-API-Key": "dev-key"}' \
SERVER_NAME="dev_api" \
python src/openapi_mcp/fastmcp_server.py

🛠️ Examples & Use Cases

Norwegian Weather API

Test with real weather data (no authentication required):

# Start weather server
source venv/bin/activate && \
OPENAPI_URL="https://api.met.no/weatherapi/locationforecast/2.0/swagger" \
SERVER_NAME="weather" \
MCP_HTTP_ENABLED="true" \
MCP_HTTP_PORT="8001" \
python src/openapi_mcp/fastmcp_server.py

Available tools:

• weather_get__compact - Weather forecast for coordinates
• weather_get__complete - Detailed weather forecast
• weather_get__status - Server status

Example usage in Claude:

• "What's the weather in Oslo tomorrow?" → Uses lat=59.9139, lon=10.7522
• "Show me detailed weather for Bergen" → Uses lat=60.3913, lon=5.3221

Pet Store API

Test with Swagger's demo API:

# Start petstore server
source venv/bin/activate && \
OPENAPI_URL="https://petstore3.swagger.io/api/v3/openapi.json" \
SERVER_NAME="petstore" \
MCP_HTTP_ENABLED="true" \
MCP_HTTP_PORT="8002" \
python src/openapi_mcp/fastmcp_server.py

Available tools:

• petstore_addPet - Add a new pet to the store
• petstore_findPetsByStatus - Find pets by status
• petstore_getPetById - Find pet by ID

🏗️ Architecture

FastMCP-Based Design

src/
├── fastmcp_server.py      # FastMCP-based main server (recommended)
├── server.py              # Legacy MCP server (fallback)
├── config.py              # Configuration management
├── auth.py                # OAuth authentication handling
├── openapi_loader.py      # OpenAPI spec loading and parsing
├── request_handler.py     # Request preparation and validation
├── schema_converter.py    # Schema conversion utilities
├── exceptions.py          # Custom exception hierarchy
└── __init__.py           # Package initialization

Key Features

✅ FastMCP Integration - Uses latest FastMCP framework
✅ Automatic Tool Registration - Converts OpenAPI operations to MCP tools
✅ Multi-Transport Support - stdio, HTTP, SSE
✅ Parameter Validation - Type conversion and validation
✅ Error Handling - Comprehensive JSON-RPC error responses
✅ Authentication - OAuth2 and username/password support

How It Works

1. Configuration Loading: Validates environment variables and server configuration.
2. OpenAPI Spec Loading: Fetches and parses OpenAPI specifications with comprehensive error handling.
3. Component Initialization: Sets up modular components with dependency injection.
4. Tool Registration: Dynamically creates MCP tools from OpenAPI operations with full metadata.
5. Resource Registration: Converts OpenAPI schemas into MCP resources with proper URIs.
6. Prompt Generation: Creates contextual usage prompts and CRUD operation examples.
7. Authentication: Handles both OAuth2 and username/password authentication with token caching and automatic renewal.
8. Request Processing: Advanced parameter parsing, type conversion, and validation.
9. Error Handling: Comprehensive exception handling with structured error responses.

sequenceDiagram
    participant LLM as LLM (Claude/GPT)
    participant MCP as OpenAPI-MCP Proxy
    participant API as External API

    Note over LLM, API: Communication Process

    LLM->>MCP: 1. Initialize (initialize)
    MCP-->>LLM: Metadata, tools, resources, and prompts

    LLM->>MCP: 2. Request tools (tools_list)
    MCP-->>LLM: Detailed list of tools, resources, and prompts

    LLM->>MCP: 3. Call tool (tools_call)

    alt With OAuth2
        MCP->>API: Request OAuth2 token
        API-->>MCP: Access Token
    end

    MCP->>API: 4. Execute API call with proper formatting
    API-->>MCP: 5. API response (JSON)

    alt Type Conversion
        MCP->>MCP: 6. Convert parameters to correct data types
    end

    MCP-->>LLM: 7. Formatted response from API

    alt Dry Run Mode
        LLM->>MCP: Call with dry_run=true
        MCP-->>LLM: Display request information without executing call
    end

Resources & Prompts

The server automatically generates comprehensive metadata to enhance AI integration:

Resources

• Schema-based Resources: Automatically derived from OpenAPI component schemas
• Structured URIs: Resources are registered with consistent URIs (e.g., /resource/{server_name}_{schema_name})
• Type Conversion: OpenAPI schemas are converted to MCP-compatible resource definitions
• Metadata Enrichment: Resources include server context and categorization tags

Prompts

• API Usage Guides: General prompts explaining available operations and their parameters
• CRUD Examples: Automatically generated examples for Create, Read, Update, Delete operations
• Contextual Guidance: Operation-specific prompts with parameter descriptions and usage patterns
• Server-specific Branding: All prompts are prefixed with server name for multi-API environments

Benefits

• Enhanced Discoverability: AI agents can better understand available API capabilities
• Usage Guidance: Prompts provide clear examples of how to use each operation
• Type Safety: Resource schemas ensure proper data structure understanding
• Context Awareness: Server-specific metadata helps with multi-API integration

📊 Performance & Production

Performance Characteristics

• Fast Startup: Initializes in ~2-3 seconds
• Low Memory: ~50MB base memory usage
• Concurrent Requests: Handles multiple API calls simultaneously
• Caching: Automatic OpenAPI spec and authentication token caching

Production Deployment

# Docker production deployment
docker-compose up -d

# Or with custom configuration
docker run -d \
  -e OPENAPI_URL="https://your-api.com/openapi.json" \
  -e SERVER_NAME="your_api" \
  -e MCP_HTTP_ENABLED="true" \
  -e MCP_HTTP_PORT="8001" \
  -p 8001:8001 \
  openapi-mcp:latest

Monitoring

• Health check endpoint: GET /health
• Metrics via structured logging
• Error tracking with JSON-RPC error codes

🔍 Troubleshooting

Common Issues

❌ RequestHandler.prepare_request() missing arguments

# Solution: Use fastmcp_server.py instead of server.py
python src/openapi_mcp/fastmcp_server.py  # ✅ Correct

❌ Claude Desktop doesn't see the tools

# Check configuration location
ls ~/Library/Application\ Support/Claude/claude_desktop_config.json

# Restart Claude Desktop after config changes

❌ Connection refused on port 8001

# Check if server is running
lsof -i :8001

# Check server logs for errors

❌ SSL/TLS errors with OpenAPI URLs

# Update certificates
pip install --upgrade certifi httpx

Testing Tools

Test server initialization:

python test_weather_oslo.py

Test with mcp-remote:

npx mcp-remote http://127.0.0.1:8001/sse

Check available tools:

curl http://127.0.0.1:8001/health

Environment Issues

Python version mismatch:

# Ensure Python 3.12+
python --version

# Recreate virtual environment if needed
rm -rf venv && python3.12 -m venv venv

Missing dependencies:

# Reinstall requirements
pip install --upgrade -r requirements.txt

🤝 Contributing

1. Fork this repository
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Commit changes: git commit -m 'Add amazing feature'
4. Push to branch: git push origin feature/amazing-feature
5. Open a Pull Request

📄 License

MIT License

Audits

Star History

If you find it useful, please give it a ⭐ on GitHub!