MCP Gateway

MCP Gateway

A generic, subclassable MCP gateway that routes requests between MCP clients and WebSocket-connected providers, allowing providers to register and execute tools asynchronously.

Category
访问服务器

README

<p align="center"> <img src="extras/images/logo.png" width="380" alt="MCP Gateway" /> </p>

<p align="center"> A generic, subclassable Model Context Protocol gateway whose tools are registered and executed asynchronously by connected providers over WebSocket. </p>

<p align="center"> <code>MCP Client</code> ⇄ Streamable HTTP ⇄ <code>Gateway</code> ⇄ WebSocket ⇄ <code>Provider</code> </p>

<p align="center"> <a href="https://github.com/mcp-gtw/mcp-gtw/actions/workflows/ci.yml"><img src="https://github.com/mcp-gtw/mcp-gtw/actions/workflows/ci.yml/badge.svg" alt="CI"></a> <a href="https://pypi.org/project/mcp-gtw/"><img src="https://img.shields.io/pypi/v/mcp-gtw.svg" alt="PyPI"></a> <a href="https://pypi.org/project/mcp-gtw/"><img src="https://img.shields.io/pypi/pyversions/mcp-gtw.svg" alt="Python"></a> <a href="LICENSE.md"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License: MIT"></a> </p>


The gateway never knows the capabilities ahead of time. A provider — typically a browser app — connects, publishes its own MCP tools, resources, prompts and more, and executes them when an MCP client calls them. The gateway only registers, publishes and routes, staying completely domain agnostic.

It is a small, installable library. You build a real application by subclassing Gateway and overriding hooks. The handlers themselves are written in JavaScript. This library never prescribes a specific frontend — the Provider SDK guide teaches the JS side generically.

📦 Install

pip install mcp-gtw
# or
uv add mcp-gtw

🚀 The smallest gateway

from mcp_gtw.gateway import Gateway

app = Gateway().create_app()
uv run python -m mcp_gtw.main

This publishes a real MCP endpoint at /mcp, a private provider WebSocket at /provider and a health check at /health. It exposes whatever capabilities the connected provider registers.

☁️ One-click deploy

The image runs production-ready as a non-root process. Deploy it to any host that keeps a persistent server (the /provider WebSocket needs one — serverless like Vercel does not work):

Deploy to Render

Render, Railway and Fly.io run the Dockerfile as-is (they inject PORT, which the gateway reads); on a VPS (Hostinger, …) use Docker Compose. Full guide, including reverse proxy and TLS: deployment.

🧩 Extending it

Subclass Gateway and override the hooks to attach your own domain logic:

from mcp_gtw.channel import Channel
from mcp_gtw.gateway import Gateway

class MyGateway(Gateway):
    mcp_server_name = "my-app"

    async def on_provider_connected(self, channel: Channel) -> None:
        ...  # a provider session just came online

    def register_routes(self, app) -> None:
        super().register_routes(app)
        ...  # add your own HTTP and WebSocket routes

app = MyGateway().create_app()

Every behaviour is a swappable strategy with a secure default — authentication, tokens, origins, expiry and the wire codec. Change one by setting a *_class attribute or injecting an instance, without touching the transport. See the Gateway library guide for every override point, Extensibility for the strategy contracts, and Auth recipes for token, username/password and client-supplied-token models.

📚 Documentation

Guide What it covers
Architecture Components, transports and request flows.
Quick start Install, run and connect an MCP client.
Gateway library The Gateway class and every override point.
Extensibility The swappable strategies, their contracts and the invariants.
Auth recipes Token, username/password and client-supplied-token models.
Configuration Every setting and environment variable.
Provider protocol The private gateway ⇄ provider message protocol.
Provider SDK Writing the JavaScript provider and registering tools.
Browser console Turn any open page into a provider from DevTools.
MCP clients Connecting Claude Code, generic clients and the Inspector.
Admin dashboard The optional monitoring dashboard and its stats API.
Security The security model, tokens, origins and hardening.
Testing Running the suite and the 100% coverage gate.
Deployment Docker, reverse proxies and scaling.

🗂️ Layout

.
├── src/mcp_gtw/     # the library
├── tests/               # unit and integration tests (100% coverage)
└── docs/                # the guides linked above

✅ Requirements

  • Python 3.12+ — tested on 3.12, 3.13 and 3.14 in CI (3.12 is the pinned local and Docker version)
  • Any MCP client (Claude Code, Cursor, the MCP Inspector, …)

💜 Support

If this project saved you time, consider supporting it: GitHub Sponsors · Ko-fi.

Made with care by Paulo Coutinho.

Licensed under MIT.

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选