gemini-cli-mcp Python Server

This directory contains the Python implementation of the gemini-cli-mcp server. It uses FastAPI to expose gemini-cli functionalities as MCP-compliant tools.

1. Features

This server exposes the following gemini-cli commands as MCP Tools:

• gemini_ask: Ask a question to the Gemini model.
• gemini_yolo: Run a complex prompt with Gemini Agent in auto-execution (--yolo) mode.
• gemini_git_commit: Generate a conventional commit message from staged changes and perform a git commit.
• gemini_git_pr: Automatically commit, push, and create a Pull Request.
• gemini_git_diff: Summarize code changes using Gemini AI.

2. Technology Stack

Category	Technology
Language	Python 3.12+
Web Framework	FastAPI
Process Exec	asyncio.subprocess
CLI Framework	Typer (via mcp-cli)
Packaging	Poetry / pyproject.toml

3. Setup

Prerequisites

• Python 3.12 or higher.
• gemini-cli installed globally and accessible in your system's PATH.
• git installed and configured.

Installation

1. Navigate to the server_py directory:

   cd server_py
   

2. Install dependencies using uv (recommended) or pip:

   uv pip install -r requirements.txt
   # or
   pip install -r requirements.txt
   

Environment Variables

The server uses environment variables for configuration. You can set these in a .env file in the project root (/path/to/project_root/.env) or directly in your environment.

• GEMINI_MODEL: Specifies the Gemini model to use (e.g., gemini-2.5-flash).
• GEMINI_ALL_FILES: Set to true to include all files in context (--all-files).
• GEMINI_SANDBOX: Set to true to enable sandbox mode (--sandbox).
• GEMINI_API_KEY: Your Gemini API key (required for Docker/server environments).
• PROJECT_ROOT: The root directory of your project (important for gemini-cli operations).
• QUERY_TIMEOUT: Timeout for gemini-cli commands in seconds.
• USE_SHELL: Set to true to execute gemini-cli commands via shell (defaults to false).
• DEBUG: Set to true to enable detailed logging to log/{date}.log.

4. Running the Server

The user can select the execution mode via a CLI flag.

• STDIO Mode: python main.py (for direct CLI interaction)
• HTTP Mode: uvicorn main:app --host 0.0.0.0 --port 8000 (for AI agent integration)

Docker

A Dockerfile is provided to build and run the server in a container.

1. Build the Image: From the project root, run:

   docker build -t gemini-cli-mcp-python -f server_py/Dockerfile .
   

2. Run the Container:

   # Using an .env file
   docker run --env-file ../.env -p 8000:8000 gemini-cli-mcp-python
   
   # Passing environment variables directly
   docker run -e GEMINI_API_KEY=your_api_key -p 8000:8000 gemini-cli-mcp-python
   

5. Packaging & Distribution

The package will be distributed on PyPI. The pyproject.toml file defines a script entry point for the gemini-cli-mcp command, which will be deployed using poetry build and twine.

CLI Usage

After installing the package via pip, you can use the CLI entry point:

$ gemini-cli-mcp

This will launch the server in STDIO mode. To run in HTTP mode, use:

$ gemini-cli-mcp --http

6. Tool Usage

The server acts as a smart wrapper around gemini-cli. It constructs and executes the appropriate gemini-cli command based on the MCP tool invocation.

For example:

• gemini_ask(question="What is AI?") translates to gemini ask --model {model} --all-files --sandbox --prompt "What is AI?"
• gemini_yolo(prompt="Do something complex.") translates to gemini agent --model {model} --all-files --sandbox --yolo --prompt "Do something complex."

7. Logging

Set the DEBUG environment variable to true to enable detailed logging to server_py/log/{YYYY-MM-DD}.log.

8. MCP Client Integration Guide

The gemini-cli-mcp server supports both HTTP and STDIO modes. Below are instructions and configuration examples for integrating as an external MCP server in clients like Cursor, Windsurf, and Claude Code.

8.1 Integration via HTTP Mode

1. Start the server

   gemini-cli-mcp --http
   # or
   uvicorn main:app --host 0.0.0.0 --port 8000
   

   • Default port is 8000.
   • Use --host 0.0.0.0 for remote access.

2. Register the MCP server in your client

   • MCP server URL: http://localhost:8000 (or your server's IP)

Cursor, Windsurf Example

// cursor: $HOME/.cursor/mcp.json
// windwurf: $HOME/.codeium/windsurf/mcp_config.json
```json
{
  "mcpServers": {
    "gemini-cli-mcp": {
      "url": "http://localhost:8000"
    }
  }
}

---

8.2 Integration via STDIO Mode

1. No need to start the server manually
   • The MCP client will launch the process and communicate via STDIO.
   • Just register the following configuration.

Cursor, Windsurf Example

// cursor: $HOME/.cursor/mcp.json
// windwurf: $HOME/.codeium/windsurf/mcp_config.json
{
  "mcpServers": {
    "gemini-cli-mcp": {
      "type": "stdio",
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/project_root/server_py/main.py"
        "run",
        "main.py"
      ],
      "env": {
        "GEMINI_MODEL": "gemini-2.5-flash",
        "PROJECT_ROOT": "/path/to/project_root"
      }
    }
  }
}

Claude Code Example

// 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
{
  "mcpServers": {
    "gemini-cli-mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/project_root/server_py/main.py"
        "run",
        "main.py"
      ],
      "env": {
        "GEMINI_API_KEY": "your_api_key",
        "GEMINI_MODEL": "gemini-2.5-flash",
        "PROJECT_ROOT": "/path/to/project_root"
      }
    }
  }
}

---

8.3 Integration via pip install

1. Install package

pip install gemini-cli-mcp

2. Register the MCP server in your client

Cursor, Windsurf Example

{
  "mcpServers": {
    "gemini-cli-mcp": {
      "type": "stdio",
      "command": "gemini-cli-mcp",
      "args": [],
      "env": {
        "GEMINI_MODEL": "gemini-2.5-flash",
        "PROJECT_ROOT": "/path/to/project_root"
      }
    }
  }
}

Claude Code Example

{
  "mcpServers": {
    "gemini-cli-mcp": {
      "command": "gemini-cli-mcp",
      "args": [],
      "env": {
        "GEMINI_API_KEY": "your_api_key",
        "GEMINI_MODEL": "gemini-2.5-flash",
        "PROJECT_ROOT": "/path/to/project_root"
      }
    }
  }
}

---

Notes:

• HTTP mode allows multiple clients to connect over the network.
• STDIO mode launches a separate process per client.
• Adjust environment variables (env) as needed for your use case.
• If the server and client are on different machines, ensure firewall/port forwarding is configured appropriately.