Gemini Bridge MCP

A Model Context Protocol (MCP) server that bridges Claude, VS Code, and other MCP clients to Google's Gemini CLI, enabling powerful AI-powered file system operations and shell command execution.

Features

• 🔗 Universal Bridge: Connect any MCP client (Claude Desktop, VS Code, Claude Code) to Gemini CLI
• 📁 File System Access: Read, write, and edit files through Gemini
• 🖥️ Shell Commands: Execute complex shell operations via Gemini
• 🎯 Model Selection: Choose from Gemini 2.5 or 3 models
• 🔐 Approval Modes: Control permissions (default, auto_edit, yolo)
• 🧪 Sandbox Mode: Secure isolated execution
• 📂 Multi-Directory: Work across multiple project directories
• 🔒 Type-Safe: Built with TypeScript for reliability

Prerequisites

• Node.js 18.0.0 or higher
• Gemini CLI installed and configured
• Google AI API Key from Google AI Studio

Quick Start

1. Install Gemini CLI

npm install -g @google/gemini-cli

2. Login to Gemini

gemini auth login

3. Verify Setup

npx gemini-bridge-mcp verify

Configuration

Claude Code (VS Code Extension)

claude mcp add gemini-bridge -- npx -y gemini-bridge-mcp

Claude Desktop

Edit your config file:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "gemini-bridge": {
      "command": "npx",
      "args": ["-y", "gemini-bridge-mcp"]
    }
  }
}

Usage

After configuration, restart your MCP client. The gemini_execute tool will be available.

Example Prompts

• "Use Gemini to create a new React component called UserProfile"
• "Ask Gemini to analyze this codebase and suggest improvements"
• "Have Gemini fix the TypeScript errors in src/utils.ts"

Available Tools

gemini_execute

Execute tasks using Gemini CLI with file system and shell access.

| Parameter | Type | Default | Description | |-----------|------|---------|-------------| | prompt | string | required | The task or instruction for Gemini CLI | | model | enum | gemini-3-pro-preview | Model: gemini-2.5-flash, gemini-2.5-pro, gemini-3-flash-preview, gemini-3-pro-preview | | approvalMode | enum | yolo | Permissions: default (confirm all), auto_edit (auto-approve edits), yolo (auto-approve all) | | outputFormat | enum | text | Output: text or json | | sandbox | boolean | false | Run in sandboxed mode for isolation | | workingDirectory | string | - | Execute from a specific directory | | includeDirectories | array | - | Include additional directories (max 5) | | debug | boolean | false | Enable verbose output |

Example

{
  "prompt": "Create a TypeScript function that validates email addresses",
  "model": "gemini-2.5-pro",
  "approvalMode": "yolo",
  "workingDirectory": "/path/to/project"
}

Development

Build from Source

git clone https://github.com/ahaydarli/gemini-bridge-mcp.git
cd gemini-bridge-mcp
npm install
npm run build

Scripts

| Script | Description | |--------|-------------| | npm run build | Compile TypeScript | | npm run dev | Run in dev mode with tsx | | npm run verify | Check system requirements | | npm run typecheck | Type check only |

Troubleshooting

"Gemini CLI not found"

npm install -g @google/gemini-cli

"Authentication error"

gemini auth login

"Command timeout"

The default timeout is 10 minutes. For longer tasks, break them into smaller operations.

Security Considerations

• Gemini CLI has full access to your file system and can execute shell commands
• Default approval mode is yolo (auto-approve all) - use default for more control
• Use sandbox mode (sandbox: true) for untrusted operations
• Only use in trusted directories with trusted prompts

License

MIT License - see the LICENSE file for details.

Links

• npm Package
• Model Context Protocol
• Gemini CLI Docs