Gemini AI MCP Server

This project integrates Google's Gemini AI models with Claude CLI via the MCP protocol.

Core Features

• MCP Protocol Integration: Seamless integration with Claude CLI through standardized MCP protocol
• Multiple Model Support: Compatible with various Gemini AI models including 2.5 Pro and 1.5 series
• JSON Response Format: Structured JSON responses for reliable parsing and integration
• Error Handling: Robust error handling with informative error messages
• Command-line Arguments: Flexible command-line interface with optional parameters
• Environment Configuration: Simple API key configuration through environment variables

Setup

1. Ensure you have the Google Generative AI Python package installed:

   pip install google-generativeai
   

2. Set your Google API key as an environment variable:

   export GOOGLE_API_KEY=your_api_key
   

3. Make sure the shell script is executable:

   chmod +x gemini.sh
   

Usage with Claude CLI

1. Configure Claude CLI to use this MCP server by adding the directory to your MCP search path

2. Use the Gemini command through Claude CLI:

   claude mcp gemini --prompt "Your prompt here"
   

3. Optionally specify a different Gemini model:

   claude mcp gemini --prompt "Your prompt here" --model "gemini-1.5-pro"
   

Testing the Server Directly

You can test the server directly without Claude CLI:

./gemini.sh gemini --prompt "Explain quantum physics simply"

Available Models

• gemini-2.5-pro-preview-03-25 (default)
• gemini-1.5-pro
• gemini-1.5-flash

MCP Integration

This server implements the MCP protocol for seamless integration with Claude CLI. The mcp.json file defines the available commands and parameters.

Architecture

gemini_server/
├── mcp.json            # MCP protocol definition
├── gemini_server.py    # Core server implementation
├── gemini.sh           # Shell wrapper script
├── genai.py            # Original implementation (reference)
├── README.md           # Documentation
└── changes.log         # Implementation changes log