┌─────────────────────────────────────────────────────────────────┐
│                                                                 │
│   Claude / Cursor / Any MCP Client                              │
│                    │                                            │
│                    ▼                                            │
│   ┌─────────────────────────────────────┐                       │
│   │     @staticpayload/gemini-mcp       │                       │
│   │   ┌─────────┐ ┌─────────┐ ┌───────┐ │                       │
│   │   │ prompt  │ │ models  │ │  raw  │ │   ◄── MCP Tools       │
│   │   └────┬────┘ └────┬────┘ └───┬───┘ │                       │
│   └────────┼───────────┼─────────┼──────┘                       │
│            │           │         │                              │
│            └───────────┼─────────┘                              │
│                        ▼                                        │
│            ┌───────────────────────┐                            │
│            │    Gemini CLI         │   ◄── Your existing auth   │
│            │    (gemini binary)    │       & configuration      │
│            └───────────────────────┘                            │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

✨ Features

🔌 Zero Configuration

Uses your existing Gemini CLI installation and authentication. No API keys to manage, no duplicate auth flows.

🚀 Universal Access

Register once, use Gemini from Claude, Cursor, Windsurf, or any MCP-compatible client.

⚡ Production Ready

Health checks, graceful shutdown, 5-minute timeouts, proper signal handling. Built for reliability.

🛠️ Three Powerful Tools

Prompt execution, model listing, and raw CLI access for advanced use cases.

🚀 Quick Start

Prerequisites

# Install Gemini CLI globally
npm install -g @google/gemini-cli

# Authenticate (run once)
gemini

Run the MCP Server

npx @staticpayload/gemini-mcp

That's it. The server starts and waits for MCP connections via stdio.

🔧 Tools

gemini_prompt

Send a prompt to Gemini and get a response.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | prompt | string | ✅ | The prompt to send | | model | string | ❌ | Model override (e.g., gemini-2.5-flash) |

gemini_models

List all available Gemini models.

No parameters required.

gemini_raw

Execute any Gemini CLI command with raw arguments.

| Parameter | Type | Required | Description | |-----------|------|----------|-------------| | args | string[] | ✅ | CLI arguments array |

Example: ["--version"] or ["-p", "Hello", "-m", "gemini-2.5-pro"]

📋 Usage

With Claude Desktop

Add to your Claude configuration (~/.claude/config.json):

{
  "mcpServers": {
    "gemini": {
      "command": "npx",
      "args": ["@staticpayload/gemini-mcp"]
    }
  }
}

With Claude CLI

claude mcp add gemini -- npx @staticpayload/gemini-mcp

With Cursor / Windsurf

Add to your MCP settings:

{
  "gemini": {
    "command": "npx",
    "args": ["@staticpayload/gemini-mcp"]
  }
}

⚙️ Configuration

Environment Variables

| Variable | Description | |----------|-------------| | GEMINI_CLI_PATH | Override the Gemini binary location | | GEMINI_API_KEY | Gemini API key (if not using OAuth) | | GOOGLE_APPLICATION_CREDENTIALS | Service account credentials path |

Inherited Configuration

The server inherits your full environment, so existing Gemini configuration works automatically:

~/.config/gemini/          ← CLI configuration
~/.gemini/settings.json    ← Gemini settings
GEMINI_* env vars          ← All Gemini environment variables
gcloud auth                ← Application default credentials

🤖 Autonomous Setup

For production deployments with Vertex AI authentication and zero interactive prompts, use the automated setup:

# 1. Copy environment template
cp .env.example .env

# 2. Edit .env with your GCP project
# PROJECT_ID=your-gcp-project-id

# 3. Authenticate with gcloud
gcloud auth application-default login

# 4. Run automated setup
./setup-gemini.sh

# 5. Use Gemini autonomously
./run-gemini.sh

Features:

• ✅ Vertex AI authentication (no API keys)
• ✅ No permission prompts for file/shell/web operations
• ✅ GA/Preview model routing
• ✅ MCP server integration
• ✅ Sandbox directory (avoids macOS permission issues)

See GEMINI_SETUP.md for complete documentation.

🔬 How It Works

┌──────────────┐     stdio      ┌─────────────────┐     spawn     ┌─────────────┐
│  MCP Client  │ ◄────────────► │  gemini-cli-mcp │ ◄───────────► │ gemini CLI  │
│  (Claude)    │   JSON-RPC     │    (Node.js)    │   child proc  │  (binary)   │
└──────────────┘                └─────────────────┘               └─────────────┘

1. MCP Client sends JSON-RPC requests over stdio
2. gemini-cli-mcp translates MCP tool calls to Gemini CLI commands
3. Gemini CLI executes with your existing auth & config
4. Response flows back through the same path

The server is a thin translation layer—all heavy lifting happens in Gemini CLI.

🏗️ Architecture

@staticpayload/gemini-mcp/
├── src/
│   └── index.js      # MCP server (single file, ~300 lines)
├── package.json      # npm package with bin entry
└── README.md

Design Principles:

• Single responsibility: translate MCP ↔ Gemini CLI
• Zero global state
• Fail fast with clear errors
• Minimal dependencies (@modelcontextprotocol/sdk, zod)

🐛 Troubleshooting

"Gemini CLI not found"

# Ensure gemini is installed and in PATH
which gemini

# Or set the path explicitly
export GEMINI_CLI_PATH=/path/to/gemini

"Auth method not set"

# Option 1: Run Gemini CLI once to authenticate
gemini

# Option 2: Set API key
export GEMINI_API_KEY=your-api-key

Server not responding

Check stderr output for health check results:

[gemini-mcp] Gemini CLI: /usr/local/bin/gemini (0.22.4)

📄 License

GPL-3.0 © 2024