Skip to main content

Backend Architecture

The MCP Gatekeeper backend is built with FastAPI and provides REST APIs, WebSocket support, and MCP server management.

Technology Stack​

  • Python 3.9+
  • FastAPI - Async web framework
  • SQLAlchemy - ORM for database operations
  • SQLite - Local database storage
  • Pydantic - Data validation
  • httpx - HTTP client for proxy

Application Structure​

app/
├── api/              # API route handlers
│   ├── agents.py
│   ├── http_servers.py
│   ├── agent_http_servers.py
│   ├── packages.py
│   ├── servers.py
│   ├── metrics.py
│   ├── logs.py
│   └── websocket.py
├── core/             # Core configuration
│   ├── config.py
│   ├── database.py
│   └── middleware.py
├── models/           # Database models
│   ├── server.py
│   ├── metric.py
│   ├── log.py
│   └── http_server.py
├── schemas/          # Pydantic schemas
├── services/         # Business logic
│   ├── mcp_client.py
│   ├── http_proxy.py
│   ├── package_manager.py
│   └── agent_config_scanner.py
└── main.py          # Application entry point

Key Components​

FastAPI Application​

  • Binds to 127.0.0.1:8000 (localhost only)
  • Supports CORS for frontend at localhost:5173
  • Auto-generated OpenAPI docs at /docs

Database Models​

  • Server: MCP server configurations
  • Metric: Performance and usage metrics
  • Log: Server execution logs
  • HttpMcpServer: HTTP server configurations
  • HttpServerMetric: HTTP proxy metrics

Services​

  • MCP Client: JSON-RPC communication with STDIO servers
  • HTTP Proxy: Transparent proxy for HTTP MCP servers
  • Package Manager: npm package installation
  • Agent Config Scanner: Detects and modifies agent configs

Authentication​

  • Token-based auth via X-MCPGK-Token header
  • Default token: dev-token-change-in-production
  • Configurable via environment variable

WebSocket​

  • Real-time metrics broadcasting
  • Endpoint: /api/ws/metrics
  • JSON message format

For API details, see API Reference.