MCP Gateway

A lightweight, security-focused FastAPI gateway for Model Context Protocol (MCP) servers. This gateway acts as an intermediary between MCP clients and servers, providing authentication, authorization, rate limiting, request validation, and audit logging.

Features

Security

• Authentication & Authorization: Token-based authentication with configurable client policies
• Request Validation: Validates requests against allowed tools and resources
• Response Sanitization: Removes sensitive information from responses
• Rate Limiting: Token bucket algorithm for request rate limiting
• Audit Logging: Comprehensive logging of all requests and responses
• Sandbox Mode: Isolates MCP server execution (planned enhancement)

Gateway Functionality

• Request Proxying: Forwards validated requests to appropriate MCP servers
• Policy Enforcement: Enforces client-specific security policies
• Error Handling: Graceful error handling with proper HTTP status codes
• CORS Support: Configurable CORS policies for web clients

Monitoring & Management

• Health Checks: Built-in health check endpoint
• Metrics: Request metrics and performance monitoring (planned)
• Admin API: Client registration and management endpoints

Quick Start

Installation

1. Clone or create the project directory:

mkdir mcpgw && cd mcpgw

2. Install dependencies:

pip install -r requirements.txt

3. Run the gateway:

python main.py

The gateway will start on http://localhost:8000

Basic Usage

1. Health Check:

curl http://localhost:8000/health

2. Register a Client (for testing):

curl -X POST http://localhost:8000/admin/register-client \
  -H "Content-Type: application/json" \
  -d '{
    "client_id": "my-client",
    "client_secret": "my-secret",
    "policy": {
      "allowed_tools": ["get_weather"],
      "allowed_resources": ["weather://*"],
      "max_requests_per_minute": 100,
      "require_auth": true,
      "sandbox_mode": true
    }
  }'

3. Make MCP Request:

curl -X POST http://localhost:8000/mcp/request \
  -H "Authorization: Bearer YOUR_TOKEN_HERE" \
  -H "Content-Type: application/json" \
  -d '{
    "server_name": "example-weather",
    "request": {
      "jsonrpc": "2.0",
      "method": "tools/list",
      "id": "1"
    }
  }'

Configuration

Security Policies

Each client can have a custom security policy:

{
  "allowed_tools": ["tool1", "tool2"],           # Allowed tool names
  "allowed_resources": ["resource://*"],         # Resource patterns (regex)
  "max_requests_per_minute": 60,                # Rate limit
  "max_request_size": 1048576,                  # Max request size (bytes)
  "require_auth": true,                         # Require authentication
  "allowed_origins": ["https://example.com"],   # CORS origins
  "sandbox_mode": true                          # Enable sandboxing
}

MCP Server Configuration

Configure MCP servers in the MCPServerManager.load_server_config() method or via configuration files:

{
  "server-name": {
    "command": "node",
    "args": ["/path/to/server/build/index.js"],
    "env": {"API_KEY": "your-key"},
    "enabled": true
  }
}

API Endpoints

Public Endpoints

• GET /health - Health check
• POST /admin/register-client - Register new client (admin only)

Authenticated Endpoints

• GET /servers - List available MCP servers
• POST /mcp/request - Proxy MCP request to server
• GET /audit/logs - Get audit logs

Security Features

Authentication

• Bearer token authentication
• Client-specific tokens with policies
• Token validation and expiration

Request Validation

• Method validation against allowed tools/resources
• Request size limits
• Input sanitization
• Policy enforcement

Rate Limiting

• Token bucket algorithm
• Per-client rate limits
• Configurable limits per security policy

Audit Logging

• All requests logged with timestamps
• Client identification
• Success/failure tracking
• Error details

Response Sanitization

• Removes sensitive fields (passwords, tokens, keys, secrets)
• Recursive sanitization of nested objects
• Configurable sensitive field patterns

Architecture

Client → Gateway → MCP Server
   ↓        ↓         ↓
Auth    Validation  Response
Rate    Logging     Sanitization
Limit   Policy      Error Handling
        Enforcement

Components

1. SecurityManager: Handles authentication, authorization, and policy enforcement
2. MCPServerManager: Manages connections and communication with MCP servers
3. RateLimitBucket: Implements token bucket rate limiting
4. FastAPI App: HTTP server with middleware and endpoints

Future Enhancements

Security

• JWT token support
• OAuth2 integration
• Role-based access control (RBAC)
• Request signing and verification
• IP whitelisting/blacklisting
• Advanced sandboxing with containers

Functionality

• WebSocket support for real-time communication
• Request/response caching
• Load balancing across multiple MCP servers
• Circuit breaker pattern for fault tolerance
• Request transformation and middleware

Monitoring

• Prometheus metrics
• Distributed tracing
• Performance monitoring
• Alert system
• Dashboard UI

Storage

• Redis backend for rate limiting
• Database storage for audit logs
• Configuration management UI
• Client management dashboard

Development

Running in Development Mode

python main.py

The server will reload automatically on code changes.

Testing

Create test clients and policies:

# Example test client
test_policy = SecurityPolicy(
    allowed_tools={"get_weather", "get_forecast"},
    allowed_resources={"weather://*"},
    max_requests_per_minute=100
)

test_client = ClientConfig(
    client_id="test-client",
    client_secret="test-secret",
    policy=test_policy
)

Adding New MCP Servers

1. Update the MCPServerManager.load_server_config() method
2. Add server configuration with command, args, and environment
3. Implement actual MCP communication in send_request() method

Security Considerations

Threat Model

The gateway protects against:

• Malicious MCP servers: Policy enforcement prevents unauthorized access
• Compromised clients: Rate limiting and validation prevent abuse
• Data exfiltration: Response sanitization removes sensitive data
• DoS attacks: Rate limiting and request size limits
• Injection attacks: Input validation and sanitization

Best Practices

1. Use strong authentication tokens
2. Implement least-privilege policies
3. Monitor audit logs regularly
4. Keep dependencies updated
5. Use HTTPS in production
6. Implement proper error handling
7. Regular security audits

License

This project is provided as an MVP example. Enhance and adapt according to your security requirements.

Contributing

This is an MVP implementation. Key areas for contribution:

• Real MCP server communication implementation
• Enhanced security features
• Performance optimizations
• Comprehensive testing
• Documentation improvements