supercorp-ai/supergateway

Run MCP stdio servers over SSE and SSE over stdio. AI gateway.

What it solves

Supergateway provides a way to bridge the gap between different communication protocols used by the Model Context Protocol (MCP). Specifically, it allows users to run MCP servers that only support standard input/output (stdio) over web-based protocols like Server-Sent Events (SSE), WebSockets (WS), and Streamable HTTP, enabling remote access, easier debugging, and integration with clients that do not support stdio.

How it works

It acts as a transport layer converter. Depending on the configuration, it can perform the following conversions:

  • stdio $\to$ SSE/WS/Streamable HTTP: Exposes a local MCP stdio server as a network service.
  • SSE/Streamable HTTP $\to$ stdio: Connects to a remote MCP server and exposes it locally as a stdio server, which is useful for integrating remote servers into local CLI environments or AI assistants like Claude Desktop and Cursor.

Who it’s for

Developers building or using MCP servers who need to expose their tools to remote clients, debug their servers using web-based tools (like the MCP Inspector), or integrate remote MCP servers into local AI agents.

Highlights

  • Multi-protocol support: Supports SSE, WebSockets, and Streamable HTTP (both stateless and stateful).
  • Flexible deployment: Can be run via npx, Docker, or as a local installation.
  • Authentication support: Includes flags for adding custom headers and OAuth2 Bearer tokens for secure remote connections.
  • Integration ready: Specifically configured for easy use with Claude Desktop and Cursor.
  • CORS and Health Checks: Includes built-in support for enabling CORS and registering health check endpoints.

Related

  • Project
    PythonOT/POT

    A Python library providing a wide range of solvers for Optimal Transport problems used in machine learning, signal processing, and image processing.

  • Project
    modelcontextprotocol/ext-apps

    A framework for building interactive UIs (charts, forms, dashboards) that render inline within AI chat clients using the Model Context Protocol.

  • Project
    deeplethe/forkd

    A microVM sandbox runtime for AI agent fan-out that uses snapshot copy-on-write to spawn hundreds of isolated, warmed runtimes in milliseconds.

  • Project
    radixark/miles

    An enterprise-grade reinforcement learning framework for large-scale model post-training that integrates SGLang for rollouts and Megatron-LM for scalable training.