punkpeye/mcp-proxy

MCP Proxy lets you take a local AI tool server that only talks through command-line-style input and output, and expose it over the web so that browser-based apps and other network clients can connect to it. This matters because many tools built for the Model Context Protocol (a standard for connecting AI assistants to external data and services) are designed to run on your local machine. Without something like this proxy, those tools can't easily be reached from a web app or a remote server. The way it works is straightforward: you launch your local tool server through the proxy, and the proxy starts listening on a web port. When a client sends an HTTP request, the proxy forwards it to your local server and passes the response back. It supports two communication patterns, streaming HTTP and server-sent events (SSE), so clients can choose either. You can run it as a one-line command pointing at any compatible server, or use it as a library inside a larger Node.js project. This is useful for developers building AI-powered applications who want to use existing MCP tools from a web interface rather than only from a terminal. For example, if you have a filesystem tool that reads local files and you want a web-based chatbot to use it, the proxy bridges that gap. It also includes a public tunnel feature, so you can share a local server with a remote colleague or test webhooks without deploying anything. A few practical touches stand out. You can add API key authentication to protect your endpoints, configure CORS so browsers on different domains can access the server, and run in a "stateless" mode that spins up a fresh server instance per request, handy for serverless platforms where you don't control persistent connections. There's also a debug mode for troubleshooting and specific guidance for Python-based servers, which can silently fail if their output buffering delays protocol messages. The project is built in TypeScript and can be used either as a quick command-line tool or as a programmable library for more custom integrations.