Skip to content
MCP server

MCP-Bridge

Gateway that exposes MCP tools through an OpenAI-compatible API for self-hosted setups

Pricing
Open source
Platforms
API, self-hosted, Web
GitHub stars
922
Last commit
1 year ago
Language
Python
License
MIT

Similar MCP servers

About MCP-Bridge

MCP-Bridge is an open-source MCP server for ai models and llm tooling. It is available on API, self-hosted and Web.

The README says MCP-Bridge sits between an OpenAI-style client and an inference engine, injecting the tool definitions from configured MCP servers into chat requests and running the resulting tool calls against those servers. A client that only speaks the OpenAI API can therefore use MCP tools without any native MCP support of its own. Configuration lives in a JSON file that lists the inference endpoint and the MCP servers to launch, and the project recommends running it with Docker.

Beyond chat completions, the project offers REST endpoints for MCP primitives and an SSE endpoint so MCP-aware clients can connect to it as a server. The README notes that vLLM has been tested and that the inference engine must support tool calls. It also flags that streaming completions are not yet implemented, that MCP resources are planned, and that the maintainers are looking for new contributors and consider the project soft deprecated since Open WebUI added native MCP support.

What it does

OpenAI-compatible chat endpoints
Chat completions, both streaming and non-streaming, are served with MCP tools available to the model.
Config-driven MCP servers
MCP servers are declared in a JSON config file, loaded from a mounted file, a URL, or an environment variable.
SSE bridge for MCP clients
Exposes the aggregated tools over SSE so MCP-aware clients can connect to the bridge as a server.
API key authentication
Optional bearer-token checks can be enabled in the config to restrict access to the server.
REST endpoints for MCP primitives
Exposes REST routes for interacting with MCP features, documented through the built-in OpenAPI docs.
MCP sampling support
The README lists MCP sampling as a working feature, routed to configured models with cost and speed weights.

Good for

  • Give a chat front end that lacks MCP support access to tools such as web fetch
  • Test whether MCP server configuration works by listing tools through the API or SSE endpoint
  • Run a local model served by vLLM or Ollama with MCP tools attached
  • Place API key authentication in front of a shared MCP tool setup

Pricing

Open source

MIT-licensed, free to self-host

MCP-Bridge FAQ

Does it support streaming responses?

Streaming chat completions with MCP are listed as working, while streaming completions without MCP are described as not yet implemented.

Which inference engines are mentioned?

The README says vLLM has been tested successfully and that ollama should also be compatible, provided it supports tool calls.

What is the recommended way to install it?

The README recommends Docker with a compose file, and also documents a manual install using uv.

Is MCP-Bridge free?

Yes. MCP-Bridge is open source, so you can use and self-host it for free.

What are the best alternatives to MCP-Bridge?

The closest alternatives to MCP-Bridge are Proxima, MewCP, Apollo MCP Server, awesome-mcp-servers and MCP.

Where this comes from

  • Description: written from the project's README.

How AIPoweredBot checks servers