Gemini MCP Server

gemini-mcp

Lightweight MCP server that exposes Google Gemini as tools for Claude Code (or any MCP client).

Use Gemini for second opinions, large-context analysis, code review, or anything where a different model perspective helps.

Tools

Quick Start

1. Get an API key

Go to Google AI Studio and create a free API key.

2. Install

Option A — Clone (recommended for Claude Code)

git clone https://github.com/PavelGuzenfeld/gemini-mcp.git ~/.claude/mcp-servers/gemini
cd ~/.claude/mcp-servers/gemini
npm install

Option B — npx (no install)

npx claude-gemini-mcp

3. Register with Claude Code

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "gemini": {
      "command": "node",
      "args": ["/home/you/.claude/mcp-servers/gemini/index.js"],
      "env": {
        "GEMINI_API_KEY": "your-key-here"
      }
    }
  }
}

Or with npx:

{
  "mcpServers": {
    "gemini": {
      "command": "npx",
      "args": ["-y", "claude-gemini-mcp"],
      "env": {
        "GEMINI_API_KEY": "your-key-here"
      }
    }
  }
}

Usage Examples

Ask a question

> Use gemini_ask to explain the difference between std::expected and std::optional

Gemini says: std::optional<T> represents a value that may or may not be present...
std::expected<T, E> additionally carries an error value when the expected value is absent...

Analyze code

> Use gemini_analyze to review this function for performance issues:
  instruction: "Find performance bottlenecks"
  content: <your code here>

Gemini says: Line 12 allocates inside the loop — move the vector outside...

Multi-turn conversation

> Use gemini_chat with messages:
  [{"role": "user", "content": "Design a REST API for a task manager"},
   {"role": "model", "content": "Here's a RESTful design..."},
   {"role": "user", "content": "Now add authentication"}]

Gemini says: Building on the previous design, add JWT-based auth...

Override model per call

> Use gemini_ask with model: "gemini-2.5-flash" to quickly summarize this error log

Environment Variables

Models

Every tool accepts an optional model parameter to override the default per-call.

Features

• Retry with exponential backoff on rate limits (429) and server errors (5xx)
• Graceful error reporting back to the MCP client (no crashes)
• Per-call model override
• Zero configuration beyond the API key

Troubleshooting

Development

git clone https://github.com/PavelGuzenfeld/gemini-mcp.git
cd gemini-mcp
npm install
npm test          # Run smoke tests
node index.js     # Start the MCP server locally

License

MIT