MCP Gateway
A production-ready unified entry point for AI agents that implements the Model Context Protocol (MCP). It provides a secure gateway with rate limiting, authentication, and observability for managing and proxying requests to multiple downstream APIs.
README
MCP Gateway
Production-ready unified entry point for AI Agents and downstream APIs, implementing the Model Context Protocol (MCP).
Quick Start (Local — 5 minutes)
Prerequisites
| Tool | Minimum Version | Check |
|---|---|---|
| Python | 3.10+ | python3 --version |
| pip | 23+ | pip3 --version |
| make | any | make --version |
| Docker (optional) | 24+ | docker --version |
macOS: Install Python via Homebrew:
brew install python@3.12Windows: Install Python from python.org, then usemakevia WSL or Git Bash.
Option A — Run with Python directly (recommended for development)
# 1. Clone / open the project folder
cd mcp-gateway
# 2. One-command setup (creates virtualenv + installs deps + copies .env)
make setup
# 3. Start the gateway
make run
The gateway starts at http://localhost:8080.
Open the interactive API docs at http://localhost:8080/docs.
Option B — Run with Docker Compose (full stack)
# Starts gateway + Redis + Prometheus + Grafana + Jaeger
make docker-up
| Service | URL |
|---|---|
| Gateway API | http://localhost:8080 |
| API Docs | http://localhost:8080/docs |
| Grafana | http://localhost:3000 (admin/admin) |
| Prometheus | http://localhost:9091 |
| Jaeger (traces) | http://localhost:16686 |
Manual Setup (step by step)
If you prefer not to use make:
# 1. Create virtual environment
python3 -m venv .venv
source .venv/bin/activate # macOS/Linux
# .venv\Scripts\activate # Windows
# 2. Install dependencies
pip install -e ".[dev]"
# 3. Copy environment config
cp .env.example .env
# 4. Create log directory
mkdir -p logs
# 5. Start the gateway
uvicorn src.main:app --host 0.0.0.0 --port 8080 --reload
Configuration
All configuration is in the .env file (copied from .env.example during setup).
Key settings
# Auth — who can call the gateway
GATEWAY_AUTH_ENABLED=true
GATEWAY_API_KEYS_FILE=config/api_keys.json # your API keys
# Upstream APIs — what the gateway proxies to
GATEWAY_UPSTREAM_CONFIG_FILE=config/upstreams.json
# Rate limiting
GATEWAY_RATE_LIMIT__DEFAULT_RPM=60 # requests per minute per identity
GATEWAY_RATE_LIMIT__BURST_ALLOWANCE=10
# Debug / dev mode
GATEWAY_DEBUG=true
GATEWAY_ENVIRONMENT=development
Add your own upstream API
Edit config/upstreams.json:
[
{
"name": "my-api",
"url": "https://api.example.com",
"description": "My downstream API",
"upstream_token": "your-bearer-token-here",
"timeout_seconds": 30,
"tags": ["internal"]
}
]
Add an API key
Edit config/api_keys.json:
[
{
"key": "my-secret-key-001",
"owner": "your-name",
"allowed_upstreams": [],
"rate_limit_rpm": -1,
"scopes": ["read", "write"]
}
]
Try It Out
Once the gateway is running, open a new terminal and try these:
Health check (no auth required)
curl http://localhost:8080/health
List configured upstreams
curl http://localhost:8080/upstreams \
-H "X-Api-Key: dev-key-alice-001"
Send an MCP initialize request
curl -X POST http://localhost:8080/mcp \
-H "Content-Type: application/json" \
-H "X-Api-Key: dev-key-alice-001" \
-d '{
"version": "1.0",
"request": {
"jsonrpc": "2.0",
"id": 1,
"method": "initialize",
"params": {
"protocolVersion": "2024-11-05",
"clientInfo": {"name": "my-agent", "version": "1.0"}
}
}
}'
Register an AI agent (get a session token)
curl -X POST http://localhost:8080/agents/register \
-H "Content-Type: application/json" \
-H "X-Api-Key: dev-key-alice-001" \
-d '{
"agent_name": "my-llm-agent",
"agent_version": "1.0.0",
"requested_upstreams": [],
"requested_scopes": ["read", "write"]
}'
Use agent token to call MCP
# Replace <token> with the token from the register response
curl -X POST http://localhost:8080/mcp \
-H "Content-Type: application/json" \
-H "X-Agent-Token: <token>" \
-d '{
"request": {
"jsonrpc": "2.0",
"id": 2,
"method": "tools/list"
}
}'
Proxy directly to an upstream
curl -X POST http://localhost:8080/mcp/notion \
-H "Content-Type: application/json" \
-H "X-Api-Key: dev-key-alice-001" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
Common Commands
make run # Start gateway (http://localhost:8080)
make run-reload # Start with hot-reload (auto-restarts on code change)
make test # Run all tests
make test-unit # Run unit tests only
make lint # Lint code with ruff
make format # Auto-format code
make typecheck # Type check with mypy
make docker-up # Start full stack (gateway + Redis + monitoring)
make docker-down # Stop Docker stack
make clean # Remove virtualenv and caches
Project Structure
mcp-gateway/
├── src/
│ ├── main.py ← Entry point
│ ├── core/
│ │ ├── app.py ← App factory + middleware wiring
│ │ ├── config.py ← All configuration (env + JSON)
│ │ └── state.py ← Shared HTTP pool + rate buckets
│ ├── protocol/
│ │ ├── mcp_types.py ← MCP message types
│ │ └── validator.py ← Request validation pipeline
│ ├── routing/
│ │ ├── mcp_router.py ← MCP method handling
│ │ └── proxy.py ← Reverse proxy with retries
│ ├── security/
│ │ ├── auth.py ← API Key + OAuth JWT + Agent auth
│ │ └── agent_registry.py ← Agent session management
│ ├── middleware/
│ │ ├── rate_limiter.py ← Token-bucket rate limiting
│ │ ├── logging_mw.py ← Structured JSON access logs
│ │ └── circuit_breaker.py ← Per-upstream circuit breaker
│ └── observability/
│ ├── telemetry.py ← OpenTelemetry + Prometheus
│ └── health.py ← /health, /readyz, /metrics
├── config/
│ ├── api_keys.json ← API credentials
│ └── upstreams.json ← Downstream API definitions
├── tests/
│ ├── unit/ ← Fast, isolated tests
│ └── integration/ ← Full stack tests
├── deploy/k8s/ ← Kubernetes manifests
├── monitoring/ ← Prometheus + Grafana configs
├── Dockerfile ← Multi-stage production image
├── docker-compose.yml ← Local dev stack
├── Makefile ← Developer commands
└── .env.example ← Config template
Authentication
The gateway supports three auth methods, all tried in order on each request:
| Method | Header | Example |
|---|---|---|
| API Key | X-Api-Key |
X-Api-Key: dev-key-alice-001 |
| Agent Token | X-Agent-Token |
X-Agent-Token: <jwt> |
| OAuth Bearer | Authorization |
Authorization: Bearer <jwt> |
Public paths (no auth needed): /health, /readyz, /metrics, /docs
Key Endpoints
| Endpoint | Method | Auth | Description |
|---|---|---|---|
/health |
GET | None | Liveness probe |
/readyz |
GET | None | Readiness + upstream health |
/metrics |
GET | None | Prometheus metrics |
/docs |
GET | None | Swagger UI |
/mcp |
POST | Required | MCP protocol endpoint |
/mcp/{upstream} |
POST | Required | Direct upstream proxy |
/agents/register |
POST | Required | Register AI agent |
/agents/sessions |
GET | Required | List active sessions |
/upstreams |
GET | Required | List configured upstreams |
/config/info |
GET | Required | Gateway configuration info |
Troubleshooting
ModuleNotFoundError: No module named 'src'
→ Make sure you're in the mcp-gateway/ directory, not a subdirectory.
→ Run pip install -e ".[dev]" from the project root.
401 Unauthorized
→ Add header -H "X-Api-Key: dev-key-alice-001" to your requests.
→ Check config/api_keys.json contains the key you're using.
Address already in use (port 8080)
→ Change the port: uvicorn src.main:app --port 8090
→ Or update GATEWAY_PORT=8090 in your .env file.
429 Too Many Requests
→ You've hit the rate limit. Wait a minute or increase GATEWAY_RATE_LIMIT__DEFAULT_RPM in .env.
Documentation
Full architecture document: docs/ARCHITECTURE.md
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。