Gemini CLI MCP Server bridges the Model Context Protocol with Google's Gemini CLI, enabling AI agents like Cursor and Claude Desktop to invoke Gemini commands directly. The server wraps locally installed `gemini-cli` functionality and exposes it as standardized MCP tools. Key features: - Ask questions and interact with Gemini AI models through conversational interfaces - Git workflow automation including commit generation and pull request management - Dual implementations available in Python (FastAPI) and Node.js - Multiple transport modes: stdio, HTTP, and Docker containerization - Distributed via PyPI (`pip install gemini-cli-mcp`) and npm (`npm install -g gemini-cli-mcp-server`) - Requires Gemini authentication via `gemini login` or `GEMINI_API_KEY` environment variable

gemini-cli-mcp

> Quickstart for End Users: > > Install via PyPI (Python): >

> pip install gemini-cli-mcp
> 

> npm install -g gemini-cli-mcp-server
> 

> Authentication Requirement: > > Before using this server, you must either: > - Log in to gemini-cli (e.g., by running gemini login) to maintain an active login session, or > - Set your Gemini API key as the GEMINI_API_KEY environment variable. > > Without authentication, the server will not be able to invoke gemini-cli commands successfully.

gemini-cli-mcp is a server that bridges the Model Context Protocol (MCP) with the locally installed gemini-cli. It allows modern AI agents, such as Cursor and Claude Desktop, to use gemini-cli's powerful features as Tools.

This server enables invoking key gemini-cli functionalities—including ask, ~~agent, commit, and pr~~ —directly from your AI agent.

1. Project Overview

This project aims to provide a seamless integration between AI agents and gemini-cli, supporting multiple languages and environments to maximize developer experience.

Goals

Architecture

flowchart LR
    A["AI Agent(Cursor, etc.)"]
    B["gemini-cli-mcp(Python or Node.js)"]
    C["gemini-cli"]
    A -- "MCP (stdio/http)" --> B
    B -- "Shell (Subprocess)" --> C
    C -- "Shell (Subprocess)" --> B
    B -- "MCP (stdio/http)" --> A

2. Implementations

This project provides separate, language-specific implementations. Please refer to the README.md file within each implementation directory for detailed setup and usage instructions.

3. MCP Tool Reference

The server exposes gemini-cli commands as MCP tools. The core logic involves wrapping gemini-cli commands based on the tool called.

Available Tools

| Tool Name | Description | Main Params | | ------------------- | ------------------------------------------------------------------------ | ------------------------------------------------------------ | | gemini_ask | Ask a question in Ask mode. | question (string) | | gemini_yolo | Run a prompt in Agent mode with auto-execution. | prompt (string) | | ~~gemini_git_commit~~ | ~~Generate a conventional commit message and perform git commit.~~ | ~~branch_name (string, optional)~~ | | ~~gemini_git_pr~~ | ~~Automatically commit, push, and create a PR.~~ | ~~commit_message, branch_name, pr_title (all optional strings)~~ | | ~~gemini_git_diff~~ | ~~Summarize code changes using Gemini AI.~~ | ~~diff_args (string, optional)~~ |

Command Translation Example

4. MCP Client Configuration & Usage

No need to start the server manually

#### Cursor, Windsurf Example

// cursor: $HOME/.cursor/mcp.json
// windwurf: $HOME/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "gemini-cli-mcp": {
      "type": "stdio",
      "command": "gemini-cli-mcp", // gemini-cli-mcp-node for node
      "env": {
        "GEMINI_MODEL": "gemini-2.5-flash",
        "PROJECT_ROOT": "/path/to/project_root"
      }
    }
  }
}

#### Claude Code Example ``json // Settings > Developer > Edit Config > claude_desktop_config.json // find command location with which gemini-cli-mcp` // MUST provide a Gemini API key to use with Claude Desktop { "mcpServer