gemini-cli
Integrates the local gemini CLI tool with Claude Code, allowing Claude to execute Gemini AI queries via the MCP protocol.
README
Gemini CLI MCP Server
A Model Context Protocol (MCP) server that integrates the local gemini CLI tool with Claude Code. This server acts as a bridge, allowing Claude Code to execute Gemini AI queries through the standardized MCP protocol.
Made 99% with Claude Code (including this line).
How It Works
This MCP server provides a simple interface between Claude Code and the Gemini CLI tool:
- Claude Code sends a prompt via MCP protocol
- The server receives the prompt through stdio JSON-RPC communication
- Server executes
gemini -p "<prompt>"as a child process - Gemini CLI response is captured and returned to Claude Code
- Claude Code displays the result to the user
Claude Code ↔ [JSON-RPC stdio] ↔ MCP Server ↔ [child_process] ↔ gemini CLI
Prerequisites
- Node.js v18+ installed
- Gemini CLI tool installed and available in your PATH
- Test with:
gemini -p "Hello world"
- Test with:
- Claude Code installed and configured
Installation
-
Clone or download this repository:
git clone <repository-url> cd gemini-mcp-server -
Install dependencies:
npm install -
Build the project:
npm run build
Claude Code Configuration
Method 1: Using Claude Code CLI (Recommended)
claude mcp add gemini-cli node /absolute/path/to/gemini-mcp-server/dist/index.js
Method 2: Manual Configuration
Edit your Claude Code configuration file (~/.claude.json):
{
"mcpServers": {
"gemini-cli": {
"type": "stdio",
"command": "node",
"args": ["/absolute/path/to/gemini-mcp-server/dist/index.js"]
}
}
}
Important: Use the absolute path to your built index.js file.
Method 3: Development Mode
For development, you can run directly with ts-node:
{
"mcpServers": {
"gemini-cli": {
"type": "stdio",
"command": "npx",
"args": ["ts-node", "/absolute/path/to/gemini-mcp-server/src/index.ts"]
}
}
}
Setup Verification
-
Restart Claude Code after configuration changes
-
Check MCP server status:
claude mcp list -
Verify in Claude Code:
- Run
/mcpcommand - Should show
gemini-cli: connected
- Run
-
Test the integration:
Hey Claude, use the gemini-cli tool to ask: "What is TypeScript?"
Available Tools
gemini_query
Executes a query using the local Gemini CLI tool.
Parameters:
prompt(string, required): The text prompt to send to Geminimodel(string, optional): Gemini model to use (defaults togemini-2.5-flash)
Example usage in Claude Code:
Please use the gemini_query tool with prompt: "Explain the difference between async and await"
# With model selection (otherwise defaults to gemini-2.5-flash)
Please use the gemini_query tool with prompt: "What is TypeScript?" and model: "gemini-1.5-flash"
Development
Project Structure
gemini-mcp-server/
├── src/
│ ├── index.ts # Main MCP server
│ ├── gemini-cli.ts # CLI wrapper
│ └── types.ts # TypeScript interfaces
├── dist/ # Compiled output
├── package.json
├── tsconfig.json
├── PLAN.md # Implementation details
└── README.md # This file
Development Commands
# Install dependencies
npm install
# Build for production
npm run build
# Development with auto-reload
npm run dev
# Run tests
npm test
# Run tests with coverage
npm run test:coverage
# Run tests in watch mode
npm run test:watch
Development Mode
For active development, you can run the server directly:
npm run dev
Then configure Claude Code to use ts-node as shown in Method 3 above.
Troubleshooting
Common Issues
-
"Connection closed" error
- Ensure you're using absolute paths in configuration
- Check that Node.js can execute the built file
- Verify the gemini CLI is in your PATH
-
"Command not found: gemini"
- Install the Gemini CLI tool
- Ensure it's in your system PATH
- Test manually:
gemini -p "test"
-
Server not appearing in /mcp
- Restart Claude Code after configuration changes
- Check the configuration file syntax
- Verify absolute paths are correct
-
Tool not available
- Run
claude mcp listto check server status - Look for connection errors in Claude Code logs
- Run
Debugging
-
Test the server manually:
node dist/index.js # Send JSON-RPC messages via stdin for testing -
Test with simulated responses (when Gemini CLI quota is exhausted):
GEMINI_TEST_MODE=true node dist/index.js # Uses simulated responses instead of calling Gemini CLI -
Check Gemini CLI directly:
# Test with default model (gemini-2.5-flash) gemini -m gemini-2.5-flash -p "Hello world" -
Verify Claude Code configuration:
claude mcp list
Testing the MCP Server
Test 1: Basic Server Test
cd /Users/zain/code/mcp/gemini-mcp-claude
node dist/index.js
Then send these JSON-RPC messages one by one:
# 1. Initialize the server
{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"test","version":"1.0.0"}}}
# 2. List available tools
{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}
# 3. Test basic query (uses gemini-2.5-flash by default)
{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"gemini_query","arguments":{"prompt":"What is 2+2?"}}}
# 4. Test with model override
{"jsonrpc":"2.0","id":4,"method":"tools/call","params":{"name":"gemini_query","arguments":{"prompt":"Hello world","model":"gemini-1.5-flash"}}}
Press Ctrl+C to exit when done.
Test 2: Test Mode (If Quota Exhausted)
cd /Users/zain/code/mcp/gemini-mcp-claude
GEMINI_TEST_MODE=true node dist/index.js
Send the same JSON-RPC messages as above. You'll get simulated responses instead of hitting the Gemini API.
Test 3: Verify Gemini CLI Directly
# Test default model
gemini -m gemini-2.5-flash -p "What is 2+2?"
# Test alternative model
gemini -m gemini-1.5-flash -p "Hello world"
Test 4: Configure in Claude Code
# Add to Claude Code
claude mcp add gemini-cli node $(pwd)/dist/index.js
# Verify configuration
claude mcp list
# Test in Claude Code
# Run /mcp command to see connection status
Expected Results
- Initialize: Returns server info with protocol version
- List tools: Shows
gemini_querytool with model parameter support - Tool calls: Return Gemini responses or proper error messages
- Test mode: Returns simulated responses immediately
- Claude Code: Shows
gemini-cli: connectedstatus
If you hit quota limits, the server will timeout after 5 minutes (or 10 seconds in test mode) and return a proper error message.
Testing
The project includes comprehensive test suites:
Running Tests
# Run all tests
npm test
# Run tests with coverage report
npm run test:coverage
# Run tests in watch mode (for development)
npm run test:watch
Test Coverage
- Unit Tests:
tests/gemini-cli.test.ts- Tests for CLI wrapper functions - Integration Tests:
tests/mcp-server.test.ts- Full MCP server testing - JSON Error Handling: Tests for robust JSON parsing
- Model Selection: Tests for default and custom model usage
- Validation: Tests for input validation and error cases
The test suite provides:
- 92%+ coverage of the CLI module
- Full MCP protocol compliance testing
- Error scenario testing (invalid JSON, timeouts, etc.)
- Both test and production mode validation
Security Notes
- The server validates input to prevent command injection
- Prompts are properly escaped when passed to the shell
- Command execution is limited by timeout (5 minutes for production, 10 seconds for test mode)
- Only stderr is used for logging (stdout reserved for MCP protocol)
Contributing
- Fork the repository
- Create a feature branch
- Make your changes
- Test with Claude Code
- Submit a pull request
License
MIT License - see LICENSE file for details.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。