Gemini MCP Server已验证已精选

What's Gemini MCP?

Gemini MCP is a free, open-source MCP server that gives Claude Google Gemini's best tricks as tools: grounded Google Search, image generation and editing (Nano Banana Pro and Nano Banana 2), video with sound (Veo 3.1), real SVG, and landing pages.

It doesn't replace Claude. It hands Claude the things Gemini happens to do well, so you stop flipping between browser tabs. Results preview inline in Claude Desktop through MCP Apps, not as a file path you have to go and open, and full-resolution files are saved to disk.

Built by Houtini. Apache-2.0.

How to use Gemini MCP

It's on npm, so there's nothing to clone or build. You need Node.js 20.19 or newer and a Gemini API key from Google AI Studio.

Add it to Claude Desktop (or any MCP client):

{
  "mcpServers": {
    "gemini": {
      "command": "npx",
      "args": ["-y", "@houtini/gemini-mcp"],
      "env": {
        "GEMINI_API_KEY": "your-api-key-here"
      }
    }
  }
}

Or in Claude Code:

claude mcp add -e GEMINI_API_KEY=your-api-key-here -s user gemini -- npx -y @houtini/gemini-mcp

Then restart Claude and try: "Use Gemini to look up this week's London forecast and turn it into an infographic." That's one grounded generate_image call, with real dates and temperatures.

Optional settings:

• GEMINI_IMAGE_OUTPUT_DIR: where images and video are saved.
• GEMINI_DEFAULT_MODEL: pick your chat model.
• GEMINI_DEFAULT_GROUNDING=false: turn search grounding off by default.

Key features of Gemini MCP

• Chat with Google Search grounding: current answers with source links, not a training-cutoff guess. Grounding is on by default, with a thinking_level for hard problems.
• Deep research: several grounded search passes on a fast Flash model, written up into one report by a Pro model. It's tuned to finish inside Claude Desktop's timeout.
• Image generation that spells: Nano Banana Pro renders legible text and goes up to 4K. With use_search=true it looks data up before it draws (prices, weather, results). Nano Banana 2 is there when you want speed and volume.
• Conversational image editing: "make the pedals red, swap to daylight" on an existing image, with multi-turn continuity.
• Video with Veo 3.1:
  • 4 to 8 second clips, at up to 4K, with synchronised audio;
  • animate a still with firstFrameImage;
  • keep a character consistent with referenceImages;
  • vertical or landscape.
• Real SVG, not a picture of one: architecture diagrams, icons, charts and infographics as editable vector markup you can commit, in four styles.
• Landing pages from a paragraph: one self-contained HTML file, with inline CSS and no dependencies.
• Image analysis: quick descriptions, or structured extraction and reasoning. Media-resolution control cuts tokens by up to 75%.
• Nine chart design systems, which you can ask for by name for data-heavy images.
• 13 tools plus built-in help (gemini_help). Transient network failures are retried automatically.

Use cases of Gemini MCP

• Ask Claude about anything that changed this week, and get a grounded answer with sources.
• Research a topic in depth and come back with a written report.
• Make blog images, infographics and social graphics, with text that's actually spelled right.
• Edit a product shot or a hero image with plain-English instructions.
• Turn a still into a short video with sound, for social or a landing page.
• Generate architecture diagrams and charts as SVG for docs and slides.
• Mock up a landing page for a product idea in one call.
• Read screenshots, PDFs and charts, and pull out structured data.