mcp-gw
Aggregates multiple MCP servers into a single HTTP endpoint with tool namespacing, dashboard, and REST API for management.
README
mcp-gw
An aggregating MCP (Model Context Protocol) gateway that runs as a Docker container. Claude Code connects to one endpoint and gets tools from all registered backend MCP servers combined.
Architecture
Claude Code
│ HTTP MCP (localhost:3000/mcp)
▼
┌─────────────────────────────┐
│ mcp-gateway │ ← Alpine Docker container
│ ┌─────────┐ ┌──────────┐ │
│ │ MCP srv │ │Dashboard │ │
│ └────┬────┘ └──────────┘ │
│ │ mcp-gw-net │
└───────┼─────────────────────┘
│
├── backend-a:8080
├── backend-b:8080
└── ...
Tools from all backends are aggregated and namespaced as backend__tool_name so Claude Code sees them in a single tool list.
Quick Start
git clone https://github.com/gpimthong/mcp-gw
cd mcp-gw
docker compose up -d
The gateway starts on port 3000.
- Dashboard: http://localhost:3000
- MCP endpoint: http://localhost:3000/mcp
- Health: http://localhost:3000/api/health
Connect Claude Code
Add to ~/.mcp.json:
{
"mcpServers": {
"gateway": {
"type": "http",
"url": "http://localhost:3000/mcp"
}
}
}
Restart Claude Code to pick up the change.
Adding Backend MCP Servers
Backend servers must be containers on the mcp-gw-net Docker network.
# Attach an existing container to the network
docker network connect mcp-gw-net <container-name>
# Or start a new backend directly on the network
docker run -d --name my-mcp --network mcp-gw-net my-mcp-image
Then register the backend via the dashboard form or the REST API:
curl -X POST http://localhost:3000/api/backends \
-H 'Content-Type: application/json' \
-d '{"name":"my-mcp","url":"http://my-mcp:8080/mcp","transport":"http","enabled":true}'
The registration is saved to config/backends.json and reconnects automatically on gateway restart.
Dashboard
The dashboard at http://localhost:3000 shows:
- Backend cards — connection status, tool count, latency, last ping
- Live request log — real-time stream of every tool call with duration and status
- Tool catalog — all tools from all backends, expandable by backend
Use the + Add Backend button to register new backends, or click Remove / Reconnect on each card.
REST API
| Method | Path | Description |
|---|---|---|
GET |
/api/health |
Status, version, uptime, tool count |
GET |
/api/backends |
List all backends with state |
POST |
/api/backends |
Add / reconnect a backend |
DELETE |
/api/backends/:name |
Remove a backend |
POST |
/api/backends/:name/reconnect |
Force reconnect |
GET |
/api/logs |
Last 200 log entries |
GET |
/api/logs/stream |
SSE live log stream |
Tool Namespacing
Backend memory with tool search_notes → exposed to Claude Code as memory__search_notes.
The double-underscore separator avoids collisions between backends. The gateway parses the prefix at call time to route to the right backend client.
Transport Support
| Backend transport | Config value |
|---|---|
| MCP Streamable HTTP | "transport": "http" |
| MCP SSE (legacy) | "transport": "sse" |
Development
Requires Node.js 22+.
npm install
npm run dev # tsx hot-reload
npm run build # compile TypeScript → dist/
Config is read from ./config/backends.json in dev mode (set CONFIG_PATH env var to override).
Versioning
Version is set in package.json. The gateway advertises it in the MCP initialize handshake and at /api/health. Git tags follow v<semver> (e.g. v0.1.0).
Project Structure
src/
index.ts — entry point
version.ts — reads version from package.json
types.ts — shared interfaces
logger.ts — circular log buffer + SSE emitter
config.ts — load/save backends.json
backend-registry.ts — MCP client connections + heartbeat
tool-aggregator.ts — namespace + route tool calls
mcp-server.ts — MCP HTTP server (session management)
dashboard.ts — Express app (REST API + SSE)
public/
index.html — dashboard SPA (no build step)
config/
backends.json — persisted backend definitions
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。