Gemini MCP Server

A Model Context Protocol (MCP) server that provides Google Gemini AI capabilities as tools. Built with Deno and TypeScript.

Features

ask_gemini: Unified AI assistant powered by Gemini 2.5 Pro with built-in Google Search and URL analysis capabilities
6 MCP Prompts: Pre-configured workflows for common tasks, optimized for Claude Desktop integration

Prerequisites

Deno (v1.40+)
Google Gemini API key (Get one here)

Installation

Option 1: Use JSR Package (Recommended)

The easiest way is to use the published JSR package directly in your Claude Desktop configuration:

# No installation needed! Use jsr:@cong/gemini-mcp directly in your Claude config

Option 2: From Source

Clone this repository:

git clone <repository-url>
cd gemini-mcp

Set your environment variables:

export GEMINI_API_KEY=your_api_key_here
export GEMINI_MODEL=gemini-2.5-pro-preview-06-05

Usage

Start the Server

# Set your environment variables first
export GEMINI_API_KEY=your_api_key_here
export GEMINI_MODEL=gemini-2.5-pro-preview-06-05

# Development mode (with watch)
deno task dev

# Production mode
deno task start

Testing with MCP Inspector

To test the server with the MCP Inspector, make sure your environment variables are set:

# Set environment variables in your shell
export GEMINI_API_KEY=your_api_key_here
export GEMINI_MODEL=gemini-2.5-pro-preview-06-05

# Install MCP Inspector if you haven't already
npm install -g @modelcontextprotocol/inspector

# Run the inspector
npx @modelcontextprotocol/inspector src/server.ts

Important: The environment variables must be set in the same shell where you run the MCP Inspector.

Run Tests

# Run all tests
deno task test

# Run tests with watch mode
deno task test:watch

Code Quality

# Format code
deno fmt

# Lint code
deno lint

MCP Tool

ask_gemini

The unified AI assistant powered by Gemini 2.5 Pro with built-in Google Search and URL analysis capabilities.

Parameters:

prompt (required): Your question or request. Include URLs directly in the text for analysis.
temperature (optional): Controls randomness and creativity (0-2, default: 0.7)
- 0 = deterministic output
- 0.2 = focused/factual responses
- 0.7 = balanced (default)
- 1.0-2.0 = creative/diverse outputs
thinking_budget (optional): Thinking budget for reasoning (128-32,768 tokens). If not provided, model automatically decides the optimal budget.

Example:

{
  "name": "ask_gemini",
  "arguments": {
    "prompt": "What are the latest developments in quantum computing? Please analyze this paper: https://arxiv.org/abs/2301.01234",
    "temperature": 0.5,
    "thinking_budget": 2048
  }
}

Capabilities:

🔍 Automatically searches the web for current information
📄 Analyzes URLs mentioned in your prompt text
🧠 Uses Gemini 2.5 Pro with thinking capabilities, urlContext and googleSearch tools
📚 Provides comprehensive, well-sourced answers with enhanced reasoning

MCP Prompts (Claude Desktop Integration)

Pre-configured workflows that appear as slash commands in Claude Desktop:

/research_analysis

Research a topic comprehensively with automatic web search and optional URL analysis.

Arguments: topic (required), urls (optional), focus_areas (optional)
Use case: Academic research, market analysis, technology investigations

/current_events

Get up-to-date information on recent developments and news.

Arguments: topic (required), time_period (optional), region (optional)
Use case: Breaking news, recent developments, current status updates

/technical_documentation

Analyze technical documentation and provide clear explanations.

Arguments: documentation_urls (required), question (required), complexity_level (optional)
Use case: API documentation, RFCs, technical specifications

/compare_sources

Compare information across multiple sources and provide analysis.

Arguments: topic (required), source_urls (required), comparison_criteria (optional)
Use case: Research validation, source comparison, bias analysis

/fact_check

Verify claims and statements with current information and reliable sources.

Arguments: claim (required), context (optional)
Use case: Information verification, claim validation, accuracy checking

/deepthink

Deep reasoning and analysis for complex problems using maximum thinking capacity.

Arguments: problem (required), context (optional), approach (optional)
Features: Uses maximum thinking budget (32,768 tokens) with balanced temperature (0.7)
Use case: Complex decision-making, strategic planning, ethical dilemmas, multi-faceted problem analysis

Project Structure

src/
├── server.ts                    # Main MCP server implementation
├── gemini-client.ts            # Google Gemini API client wrapper
├── tools/
│   └── ask-gemini.ts           # Unified AI tool (Gemini 2.5 Pro + Search + URLs)
└── prompts/                    # MCP prompts for Claude Desktop
    ├── research-analysis.ts    # Research workflow
    ├── current-events.ts       # News and current events
    ├── technical-documentation.ts # Technical docs analysis
    ├── compare-sources.ts      # Source comparison
    ├── fact-check.ts           # Fact verification
    └── deepthink.ts            # Deep reasoning for complex problems
tests/
├── server_test.ts              # Server and prompt tests
├── gemini_client_test.ts       # API client tests
├── gemini_client_enhanced_test.ts # Enhanced client tests
├── prompts_test.ts             # Prompt logic tests
└── tools/
    └── ask_gemini_test.ts      # Tool tests

Dependencies

@modelcontextprotocol/sdk - MCP TypeScript SDK
@google/generative-ai - Google Gemini API client
zod - Runtime type validation
@std/assert - Deno standard library assertions

Development

This project uses Deno with TypeScript. Key development commands:

deno task dev - Start development server with watch mode
deno task test - Run test suite
deno cache src/server.ts - Cache dependencies
deno fmt - Format code
deno lint - Lint code

Troubleshooting

Environment Variable Issues

If you get environment variable errors:

Verify your environment variables are set:
```
echo $GEMINI_API_KEY
echo $GEMINI_MODEL
```

For MCP Inspector testing, ensure both variables are set in the same terminal:

export GEMINI_API_KEY=your_api_key_here
export GEMINI_MODEL=gemini-2.5-pro-preview-06-05
npx @modelcontextprotocol/inspector src/server.ts

Check the server logs: When the server starts, it will show (API Key: configured) to confirm your key is loaded.

Support

Claude Desktop Configuration

To use this MCP server with Claude Desktop, add it to your Claude configuration:

macOS/Linux/Windows

Edit your Claude configuration file:

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

{
  "mcpServers": {
    "gemini": {
      "command": "deno",
      "args": [
        "run",
        "--allow-net",
        "--allow-env",
        "jsr:@cong/gemini-mcp"
      ],
      "env": {
        "GEMINI_API_KEY": "your_api_key_here",
        "GEMINI_MODEL": "gemini-2.5-pro-preview-06-05"
      }
    }
  }
}

Local Development

If you're running from source code:

{
  "mcpServers": {
    "gemini": {
      "command": "deno",
      "args": [
        "run",
        "--allow-net",
        "--allow-env",
        "src/server.ts"
      ],
      "env": {
        "GEMINI_API_KEY": "your_api_key_here",
        "GEMINI_MODEL": "gemini-2.5-pro-preview-06-05"
      }
    }
  }
}

After configuration, restart Claude Desktop. The MCP prompts will appear as slash commands:

/research_analysis - Comprehensive research workflows
/current_events - Latest news and developments
/technical_documentation - Technical docs analysis
/compare_sources - Multi-source comparison
/fact_check - Claim verification
/deepthink - Deep reasoning for complex problems

Contributing

Fork the repository
Create a feature branch
Make your changes
Run tests: deno task test
Format code: deno fmt
Submit a pull request

License

MIT License - see LICENSE file for details.

Gemini