MCP-LLM Bridge CLI
A TypeScript terminal bridge: a local Ollama model calls tools across several MCP servers at once (files, email, search) based on the request text
- 984 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Picks and runs the connected MCP servers' tools based on the request text itself, with no separate confirmation before the call
- The example setup connects a Gmail and Google Drive server with access to the account's mail and files after a one-time authorization
- The filesystem MCP server can read and write files in the allowed directory at the local model's own choice
Install
Manual install
git clone https://github.com/patruff/ollama-mcp-bridge && cd ollama-mcp-bridge && npm install && npm run startBefore running, pull a model in Ollama (for example ollama pull qwen2.5-coder:7b-instruct) and the needed MCP servers, then add them to bridge_config.json.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Picks and runs the connected MCP servers' tools based on the request text itself, with no separate confirmation before the call
- The example setup connects a Gmail and Google Drive server with access to the account's mail and files after a one-time authorization
- The filesystem MCP server can read and write files in the allowed directory at the local model's own choice
This is third-party code. Review the repository files before installing.
What it does
MCP-LLM Bridge is started with `npm run start` and opens an interactive terminal where a local Ollama model answers the user's requests and decides itself which connected MCP server to call. The bridge registers each server's tools from `bridge_config.json` (filesystem, Brave search, GitHub, Gmail and Google Drive, memory, Flux image generation) and figures out which tool fits a request by keywords and context in the phrase: email operations are found by email addresses, Google Drive operations by words about files and folders, and search queries are routed to the matching search tool. Once a tool is chosen, the bridge gets a structured call from the model, validates it and passes it to the right MCP server over JSON-RPC, then returns the server's response to the model for a final text reply to the user. The README's example configures the bridge with a Qwen 2.5 7B model through Ollama, reaching the `@modelcontextprotocol/server-filesystem`, `@modelcontextprotocol/server-brave-search`, `@modelcontextprotocol/server-github`, `@modelcontextprotocol/server-memory` servers and the author's own `@patruff/server-flux` and `@patruff/server-gmail-drive` servers, but the server set and model change with `bridge_config.json`.
Who it is for. For anyone who wants to give a local Ollama model the same MCP tools Claude has access to — files, search, email, GitHub — without calling a cloud model.
Good fit when
- You want to connect a local Ollama model to several MCP servers at once without a cloud provider
- You want a terminal assistant that decides itself which connected tool to call based on the request text
- You have your own MCP servers (like the author's Flux and Gmail/Drive servers) and want a ready bridge to register them
- You want a working example of a local model with access to the filesystem, search and email in one config file
Not a fit when
- You want a ready interface that confirms each action before it runs: the README describes no confirmation step, tool detection and the call happen automatically
- You need up-to-date support for the latest MCP server versions: there have been no new commits since April 2025, and dependencies like @modelcontextprotocol/server-* may have moved the protocol on
- You need an official mail or GitHub integration: `@patruff/server-gmail-drive` and GitHub access via a personal token are the author's own servers, separate from the MCP project's official ones
- The author states no test coverage for tool routing outside the README's own examples (email, search, files): check the result by hand for an unusual request phrasing
Example request
Search the web for the latest TypeScript news and save a short summary to notes/typescript.mdLimitations
License is MIT.
- There have been no new commits since April 2025: with roughly a thousand stars that's not a reason to skip the project, but compatibility with current MCP server and Ollama versions needs checking yourself.
- Tool detection from the request text runs on keywords and phrase context (email addresses, words about files and folders), not a structured choice by the model among tool descriptions.
- The example's Gmail and Google Drive servers need separate authentication via a command like `node path/to/gmail-drive/index.js auth`, and the Flux server needs a `REPLICATE_API_TOKEN` key.
- The README's Windows install path is an absolute path like C:\Users\<name>\AppData\Roaming\npm\node_modules\..., which will differ on another system.
- The server set, model and its parameters are set in `bridge_config.json`: without editing that file, the bridge only connects to the servers and model from the author's example.
How to disable. Stop the process with the `quit` command inside the bridge's terminal or an ordinary process kill, then uninstall the MCP servers separately (`npm uninstall -g @modelcontextprotocol/server-filesystem` and similarly for the rest) and remove the repository folder itself.
FAQ
Does MCP-LLM Bridge need a cloud Anthropic or OpenAI key?
No, the bridge is built for a local model through Ollama, Qwen 2.5 7B in the README's example. Keys are only needed for individual MCP servers — Brave Search, GitHub and Flux through Replicate — not for the language model itself.
How does the bridge pick which MCP server to call?
By keywords and context in the request text: email operations are detected from email addresses in the text, Google Drive operations from words about files and folders, and search queries are routed to the matching search tool.
Does the project support current MCP server versions?
There have been no commits in the repository since April 2025, so compatibility with current @modelcontextprotocol/server-* package versions is worth checking before use. The code's structure is still built for any server that implements the MCP protocol.
Can I connect my own MCP server instead of the README's examples?
Yes, the server set is defined in bridge_config.json as an ordinary MCP configuration with a start command and a server path. The author's example uses the author's own @patruff/server-flux and @patruff/server-gmail-drive servers, but any server that answers over JSON-RPC will work.
Does Gmail and Google Drive need separate authorization?
Yes, the example's gmail-drive server needs a separate authentication step via node path/to/gmail-drive/index.js auth before first use. Without that step, the mail and drive tools stay unavailable to the bridge.
Discussion
No questions or reviews yet.
Related
CLI toolsA terminal MCP client for local Ollama models: MCP tools, prompts and resources, an agent mode and model switching
A CLI to run LLM prompts and act as an MCP client from the terminal, with any MCP server and any model provider
A CLI for any MCP server: list and call tools, resources and prompts, run mock and proxy modes, guard access and manage agent configs
An MCP server that lets Claude Code, Codex and other agents call Gemini, GPT, Grok and local models for review, debugging and consensus
oterm is a terminal chat for Ollama, OpenAI, Anthropic and other providers: SQLite-stored sessions, MCP servers, built-in and custom tools
A universal MCP CLI client from Apify: persistent sessions, OAuth, JSON output for scripts and shell access to any MCP server for agents
README in short
The repository description is short: a general paragraph on the bridge's purpose, a list of supported server capabilities (filesystem, Brave search, GitHub, Google Drive and Gmail, memory, Flux image generation) and a 'Current Setup' section naming the author's example model and servers. An architecture section with four components (bridge, LLM client, MCP client, tool router) and a step-by-step install and credential-setup guide follow.