nchan-mcp-transport

# 🚀 Nchan MCP Transport > A high-performance WebSocket/SSE transport layer & gateway for **Anthropic's MCP (Model Context Protocol)** — powered by Nginx, Nchan, and FastAPI. > For building **real-time, scalable AI integrations** with Claude and other LLM agents. --- ## ✨ What is this? **Nchan MCP Transport** provides a **real-time API gateway** for MCP clients (like Claude) to talk to your tools and services over: - 🧵 **WebSocket** or **Server-Sent Events (SSE)** - ⚡️ **Streamable HTTP** compatible - 🧠 Powered by Nginx + Nchan for **low-latency pub/sub** - 🛠 Integrates with FastAPI for backend logic and OpenAPI tooling > ✅ Ideal for AI developers building **Claude plugins**, **LLM agents**, or integrating **external APIs** into Claude via MCP. --- ## 🧩 Key Features | Feature | Description | |----------------------------------|-----------------------------------------------------------------------------| | 🔄 **Dual Protocol Support** | Seamlessly supports **WebSocket** and **SSE** with automatic detection | | 🚀 **High Performance Pub/Sub** | Built on **Nginx + Nchan**, handles thousands of concurrent connections | | 🔌 **MCP-Compliant Transport** | Fully implements **Model Context Protocol** (JSON-RPC 2.0) | | 🧰 **OpenAPI Integration** | Auto-generate MCP tools from any OpenAPI spec | | 🪝 **Tool / Resource System** | Use Python decorators to register tools and resources | | 📡 **Asynchronous Execution** | Background task queue + live progress updates via push notifications | | 🧱 **Dockerized Deployment** | Easily spin up with Docker Compose | --- ## 🧠 Why Use This? MCP lets AI assistants like **Claude** talk to external tools. But: - Native MCP is **HTTP+SSE**, which struggles with **long tasks**, **network instability**, and **high concurrency** - WebSockets aren’t natively supported by Claude — this project **bridges the gap** - Server-side logic in pure Python (like `FastMCP`) may **not scale under load** ✅ **Nchan MCP Transport** gives you: - Web-scale performance (Nginx/Nchan) - FastAPI-powered backend for tools - Real-time event delivery to Claude clients - Plug-and-play OpenAPI to Claude integration --- ## 🚀 Quickstart ### 📦 1. Install server SDK ```bash pip install httmcp ``` ### 🧪 2. Run demo in Docker ```bash git clone https://github.com/yourusername/nchan-mcp-transport.git cd nchan-mcp-transport docker-compose up -d ``` ### 🛠 3. Define your tool ```python @server.tool() async def search_docs(query: str) -> str: return f"Searching for {query}..." ``` ### 🧬 4. Expose OpenAPI service (optional) ```python openapi_server = await OpenAPIMCP.from_openapi("https://example.com/openapi.json", publish_server="http://nchan:80") app.include_router(openapi_server.router) ``` ### 🖥️ 5. One-Click GPTs Actions to MCP Deployment HTTMCP provides a powerful CLI for instant deployment of GPTs Actions to MCP servers: ```bash # Installation pip install httmcp[cli] # One-click deployment from GPTs Actions OpenAPI spec python -m httmcp -f gpt_actions_openapi.json -p http://nchan:80 ``` --- ## 📚 Use Cases - Claude plugin server over WebSocket/SSE - Real-time LLM agent backend (LangChain/AutoGen style) - Connect Claude to internal APIs (via OpenAPI) - High-performance tool/service bridge for MCP --- ## 🔒 Requirements - Nginx with Nchan module (pre-installed in Docker image) - Python 3.9+ - Docker / Docker Compose --- ## 🛠 Tech Stack - 🧩 **Nginx + Nchan** – persistent connection management & pub/sub - ⚙️ **FastAPI** – backend logic & JSON-RPC routing - 🐍 **HTTMCP SDK** – full MCP protocol implementation - 🐳 **Docker** – deployment ready --- ## 📎 Keywords `mcp transport`, `nchan websocket`, `sse for anthropic`, `mcp jsonrpc gateway`, `claude plugin backend`, `streamable http`, `real-time ai api gateway`, `fastapi websocket mcp`, `mcp pubsub`, `mcp openapi bridge` --- ## 🤝 Contributing Pull requests are welcome! File issues if you’d like to help improve: - Performance - Deployment - SDK integrations --- ## 📄 License MIT License