Overview

is a lightweight gateway written in Go that bridges the traditional Model Context Protocol (MCP) server with modern web‑based clients. By converting a standard stdio MCP interface into a Server‑Sent Events (SSE) endpoint, it enables continuous, real‑time data streams over HTTP. This solves the problem of integrating MCP servers—normally designed for command‑line or terminal use—into web browsers, single‑page applications, or AI agents that communicate over HTTP/HTTPS.

The gateway accepts a running MCP server (e.g., ) as its backend and exposes two primary HTTP paths: one for streaming SSE messages ( by default) and another for sending commands or requests back to the MCP server (). Developers can specify custom ports, base URLs, and even inject OAuth2 bearer tokens for secure access. The result is a stateless, scalable endpoint that can be deployed behind load balancers or API gateways, making it ideal for production environments where multiple AI assistants need to consume the same MCP service concurrently.

Key capabilities include:

• Real‑time streaming: SSE delivers incremental updates (e.g., token streams, progress notifications) without the overhead of WebSocket or polling.
• Simple HTTP integration: Any client that can issue GET/POST requests and handle SSE streams can interact with the MCP server, broadening the range of supported platforms.
• Security hooks: Optional bearer token support allows seamless integration with existing OAuth2 infrastructure, ensuring that only authorized agents can subscribe to or send messages.
• Configurability: Port, base URL, and path options provide flexibility to fit into diverse deployment topologies.

Typical use cases involve web dashboards that display live model outputs, browser‑based AI assistants that need to fetch data from a filesystem server, or microservices that orchestrate multiple MCP tools over HTTP. By exposing the MCP as an SSE endpoint, turns a terminal‑centric protocol into a web‑friendly API, dramatically simplifying the developer experience when building AI workflows that span both server‑side models and client‑side interfaces.