Ollama MCP Bridge

Ollama MCP Bridge is an MCP server that provides an API proxy layer between local or cloud Ollama instances and external Model Context Protocol servers. It allows software developers, researchers, and automation engineers using local language models to equip Ollama with external tool execution capabilities. The service loads configured MCP servers on startup, gathering tools across local process transports (stdio), StreamableHTTP, and SSE connections. When a client application sends a chat completion request to the bridge's /api/chat endpoint, the service injects all available MCP tool definitions into the Ollama payload. If the model invokes one or more tools, the bridge executes those calls against the target MCP servers, returns the outputs to Ollama, and handles multi-round execution loops until completion. All standard Ollama endpoints other than the chat endpoint are proxied directly to the upstream server without modification, preserving native compatibility with existing Ollama clients and frontends while providing real-time streaming and thinking-mode output.

Category: AI & LLM Tooling

Tags: bridge, llm, local-models, ollama

Visit Ollama MCP Bridge

How to install and configure Ollama MCP Bridge

1. Ensure you have Python 3.10.15 or newer and a running Ollama server instance. 2. Install the bridge package using uvx by running uvx ollama-mcp-bridge or install it with pip using pip install --upgrade ollama-mcp-bridge. 3. Create an mcp-config.json file defining your target MCP servers using stdio, HTTP, or SSE transport formats. 4. Run the bridge specifying your configuration file and target Ollama instance: ollama-mcp-bridge --config mcp-config.json --ollama-url http://localhost:11434. 5. Alternatively, launch the pre-built container using Docker: docker run -p 8000:8000 -e OLLAMA_URL=http://host.docker.internal:11434 -v "$PWD/mcp-config.json:/mcp-config.json" ghcr.io/jonigl/ollama-mcp-bridge:latest. 6. Direct your AI application or frontend client to use http://localhost:8000 as the Ollama host URL.

What you can do with Ollama MCP Bridge

• Connecting local Ollama models to stdio-based CLI developer tools and code inspection utilities. - Executing multi-round autonomous tool loops with local LLMs until a multi-step task completes. - Providing web search or live weather data to local models via external HTTP or SSE servers. - Filtering specific tools per MCP server using allowlists or denylists to restrict model capabilities. - Running automated tool calling locally inside Docker environments with multi-architecture image support.

Key facts

• Open Source
• https://github.com/jonigl/ollama-mcp-bridge
• AI & LLM Tooling, Developer Tools & Code Intelligence
• bridge, llm, local-models, ollama

Part of MCP Servers

What can Ollama MCP Bridge do?

Ollama MCP Bridge acts as a proxy in front of an Ollama server. It aggregates tools from multiple Model Context Protocol servers configured over stdio, HTTP, or SSE and exposes them to Ollama models during /api/chat requests. It handles automatic multi-round tool execution loops, tool filtering, variable expansion, response streaming, and direct proxying for non-chat endpoints.

How do I install Ollama MCP Bridge?

You can run Ollama MCP Bridge instantly without manual package setup by using uvx with uvx ollama-mcp-bridge. Alternatively, install it through pip using pip install --upgrade ollama-mcp-bridge, run it via Docker Compose using docker compose up, or deploy the official multi-arch container image directly from GitHub Container Registry.

Which MCP clients work with Ollama MCP Bridge?

Ollama MCP Bridge acts as a middleware client to other MCP servers while exposing an Ollama-compatible API. Any client application, frontend, or library that communicates with the standard Ollama API can connect to the bridge by changing the target base URL to the bridge's host address.

Is Ollama MCP Bridge open source?

Yes, Ollama MCP Bridge is an open source project distributed under the MIT license. The source code, issue tracker, and contribution guidelines are publicly accessible on GitHub.