mcp-ask-gemini

An MCP server that enables LLMs to consult with Google's Gemini models for additional insights, analysis, or alternative perspectives. This allows AI assistants to leverage multiple LLM capabilities by querying Gemini models through the official @google/generative-ai SDK.

Use Cases

This MCP server enables powerful LLM-to-LLM consultation scenarios:

• Second Opinion: Get alternative perspectives on complex problems
• Specialized Analysis: Leverage Gemini's strengths for specific domains
• Cross-Model Validation: Compare responses between different AI models
• Enhanced Reasoning: Use Gemini as a reasoning partner for complex tasks
• Research Assistance: Consult Gemini for additional insights during research

Features

• Executable via npx (no install required)
• Accepts API key, model name, and optional system prompt
• Temperature, topP, maxTokens, and stripThinkingTags with sensible defaults
• Implements exactly one MCP tool: ask_gemini
• Robust error handling for API/auth/network/rate limit issues
• TypeScript, MCP stdio transport

Installation

You can run directly with npx:

• npx mcp-ask-gemini --api-key $GOOGLE_API_KEY --model models/gemini-2.5-flash-lite

Or install locally:

• npm install mcp-ask-gemini
• npx mcp-ask-gemini --api-key $GOOGLE_API_KEY --model models/gemini-2.5-flash-lite

Configuration

You can pass configuration via CLI flags or environment variables. Defaults live in config/defaults.json, and everything there is user-overridable.

Required:

• --api-key or env API_KEY
• --model or env MODEL (e.g., models/gemini-2.5-flash-lite, models/gemini-2.5-pro). Defaults to config/defaults.json if not provided.

Optional:

• --system-prompt or env SYSTEM_PROMPT (defaults to config/defaults.json)
• --temperature or env TEMPERATURE (number; defaults to config/defaults.json)
• --topP or env TOPP (number; defaults to config/defaults.json)
• --maxTokens or env MAXTOKENS (number; defaults to config/defaults.json)
• --stripThinkingTags or env STRIPTHINKINGTAGS (boolean; defaults to config/defaults.json). Strips ... and ... blocks.

Examples:

• npx mcp-ask-gemini --api-key $GOOGLE_API_KEY --model gemini-1.5-pro
• API_KEY=$GOOGLE_API_KEY MODEL=models/gemini-2.5-flash-lite npx mcp-ask-gemini
• API_KEY=$GOOGLE_API_KEY MODEL=models/gemini-2.5-flash-lite SYSTEM_PROMPT="You are a helpful assistant." npx mcp-ask-gemini
• API_KEY=$GOOGLE_API_KEY npx mcp-ask-gemini (uses default model and generation settings)

MCP Tool: ask_gemini

The server exposes a single tool that allows LLMs to consult with Gemini models:

• name: ask_gemini
• input: { prompt: string }
• output: Gemini model text response (or an error message with isError=true)

JSON Schema Examples

Tool Request:

{
  "method": "tools/call",
  "params": {
    "name": "ask_gemini",
    "arguments": {
      "prompt": "What are the key differences between transformer and RNN architectures?"
    }
  }
}

Successful Response:

{
  "content": [
    {
      "type": "text",
      "text": "Transformers and RNNs differ in several key ways:\n\n1. **Architecture**: Transformers use self-attention mechanisms while RNNs process sequences step-by-step...\n\n2. **Parallelization**: Transformers can process all tokens simultaneously, while RNNs must process sequentially..."
    }
  ]
}

Error Response:

{
  "content": [
    {
      "type": "text",
      "text": "Authentication error with Google Gemini API. Please verify your API key. Details: Invalid API key provided"
    }
  ],
  "isError": true
}

Using with MCP clients

This server uses stdio transport, so configure your MCP client to launch the command and connect via stdio.

Claude Desktop Configuration

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "ask-gemini": {
      "command": "npx",
      "args": [
        "mcp-ask-gemini",
        "--api-key",
        "your-google-api-key-here",
        "--model",
        "gemini-1.5-pro"
      ]
    }
  }
}

MCP Inspector Configuration

• Command: npx
• Args: ["mcp-ask-gemini", "--api-key", "$GOOGLE_API_KEY", "--model", "models/gemini-2.5-flash-lite"]

Environment Variable Configuration

{
  "mcpServers": {
    "ask-gemini": {
      "command": "npx",
      "args": [
        "mcp-ask-gemini"
      ],
      "env": {
        "API_KEY": "your-google-api-key-here",
        "MODEL": "gemini-1.5-pro",
        "SYSTEM_PROMPT": "You are a helpful research assistant.",
        "TEMPERATURE": "0.7"
      }
    }
  }
}

Error Handling

The server returns useful messages for:

• Invalid API keys (authentication errors)
• Invalid or unavailable model names
• Rate limiting / quota exceeded
• Network / connectivity issues

Development

• npm install
• npm run build
• node dist/cli.js --api-key $GOOGLE_API_KEY --model models/gemini-2.5-flash-lite

Environment Variables

• API_KEY: Google Gemini API key
• MODEL: Gemini model name
• SYSTEM_PROMPT: Optional system prompt
• TEMPERATURE: Optional number override
• TOPP: Optional number override
• MAXTOKENS: Optional number override
• STRIPTHINKINGTAGS: Optional boolean override (1/0, true/false, yes/no)

License

MIT