Gemini CLI Bridge

Delegate tasks to Google's Gemini CLI from any context.

When to Use

• Google Search Grounding: Gemini has native Google Search integration
• Alternative Perspective: Get a different AI's take on a problem
• Large Context: Gemini 2.5 Pro has a 1M token context window
• Parallel Work: Have Gemini work on a subtask while you continue
• Code Review: Cross-AI code review for better coverage

Key Differences from Claude Code

Usage Patterns

Via MCP Server (Recommended)

If the gemini-cli-mcp server is configured, use the MCP tools:

Use gemini_prompt to ask Gemini: "What are the latest developments in WebGPU?"

Tools available:

• gemini_prompt - Send a prompt to Gemini CLI
• gemini_prompt_with_input - Send prompt with piped content
• gemini_search - Use Google Search grounding
• gemini_version - Get version info

Via CLI (Direct)

Execute Gemini CLI directly:

# Single prompt
gemini -p "Explain this error message" --output-format json

# Pipe input
cat error.log | gemini -p "Analyze these errors"

# With Google Search
gemini -p "What's the current price of Bitcoin?"

# YOLO mode (auto-approve all actions)
gemini -p "Fix all linting errors" --yolo

Via Rust Library

use gemini_cli_rs::GeminiCliClient;

let client = GeminiCliClient::new()?;
let response = client.prompt("Analyze this codebase").await?;
println!("{}", response.text());

Configuration

MCP Server Setup

Add to Claude Desktop's claude_desktop_config.json:

{
  "mcpServers": {
    "gemini": {
      "command": "npx",
      "args": ["@google/gemini-cli"]
    }
  }
}

Environment Variables

Best Practices

1. Specify Working Directory: Always set working_dir for file operations
2. Use YOLO Sparingly: Only enable for trusted, safe operations
3. Leverage Search: Gemini excels at queries needing real-time info
4. Check Token Stats: Monitor cached tokens for cost optimization
5. Sandbox Mode: Use sandbox: true for safer execution

Example Workflows

Current Events Research

I'll use Gemini for real-time search:

[gemini_search]
query: "What companies announced layoffs this week?"

Cross-AI Code Review

Let me get Gemini's perspective on this code:

[gemini_prompt_with_input]
prompt: "Review this code for bugs and suggest improvements"
stdin_input: "<paste code here>"

Large File Analysis

Gemini can handle larger context:

[gemini_prompt]
prompt: "Analyze the architecture of this entire codebase"
working_dir: "/home/user/large-project"

Parallel Documentation

[gemini_prompt]
prompt: "Generate API documentation for all public functions in src/"
working_dir: "/home/user/project"
yolo_mode: true

Error Handling

• Not Found: Install with npm install -g @google/gemini-cli
• Auth Failed: Run gemini interactively to authenticate with Google
• Rate Limit: Free tier allows 60 req/min, 1000 req/day
• Timeout: Increase timeout_secs for complex tasks

Authentication

Gemini CLI supports multiple auth methods:

1. Personal Google Account (recommended for personal use)

   • Run gemini and select "Login with Google"
   • 1000 requests/day free

2. Google AI Studio API Key

   export GOOGLE_API_KEY="your-key"

3. Vertex AI (for enterprise)

   export GOOGLE_API_KEY="your-key"
   export GOOGLE_GENAI_USE_VERTEXAI=true