MCP Gemini

An MCP server that lets AI assistants use Google's Gemini models. Generate text, analyze images, review code, and more — with support for multi-turn conversations and web-grounded answers.

Works with Claude Desktop, Claude Code, or any MCP-compatible client.

---

Use Cases

• Get a second opinion on code reviews
• Analyze screenshots or diagrams
• Summarize or translate long documents
• Have multi-turn conversations with Gemini for brainstorming
• Get up-to-date answers with Google Search grounding
• Generate structured JSON for use in other tools

---

Setup

Requires Node.js 18+.

1. Get a Gemini API Key

Go to Google AI Studio and create an API key.

2. Configure Your MCP Client

Add the server to your MCP client's configuration. There are two ways to do this:

Option A: Run directly from GitHub (no install needed)

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

Option B: Clone and run locally

git clone https://github.com/danjam/mcp-gemini.git
cd mcp-gemini
npm install

Then point your MCP client at the local build:

{
  "mcpServers": {
    "gemini": {
      "command": "node",
      "args": ["/path/to/mcp-gemini/dist/index.js"],
      "env": {
        "GEMINI_API_KEY": "your-api-key-here"
      }
    }
  }
}

Replace /path/to/mcp-gemini with the actual path where you cloned the project.

Changing the Default Model

By default, the server uses gemini-flash-latest. To change this, add GEMINI_DEFAULT_MODEL to your environment:

"env": {
  "GEMINI_API_KEY": "your-api-key-here",
  "GEMINI_DEFAULT_MODEL": "gemini-pro-latest"
}

---

What You Can Do

Generate Text

Ask Gemini to write, answer questions, summarize, translate, or anything else text-based.

Tool: generate_text

Parameter	Type	Required	Description
prompt	string	Yes	What you want Gemini to do
model	string	No	Which model to use
systemInstruction	string	No	Set the tone or role (e.g. "You are a helpful tutor")
temperature	number	No	How creative the response is (0 = focused, 2 = creative). Default: 0.7
maxTokens	number	No	Limit the response length. Default: 2048
grounding	boolean	No	Let Gemini search the web for up-to-date answers
conversationId	string	No	Continue a multi-turn conversation (see below)
jsonMode	boolean	No	Get the response as structured JSON
jsonSchema	object	No	Define the exact shape of the JSON response
safetySettings	array	No	Adjust content filtering thresholds

Analyze Images

Send an image and ask questions about it — describe what's in a photo, read text from a screenshot, identify objects, and more. Provide either imageUrl or imageBase64.

Tool: analyze_image

Parameter	Type	Required	Description
prompt	string	Yes	Your question about the image
imageUrl	string	No	URL of the image to analyze
imageBase64	string	No	Base64-encoded image (with or without data URI prefix)
model	string	No	Which model to use

List Models

See all available Gemini models. No parameters needed.

Tool: list_models

Code Review

Review a code diff for bugs, style issues, security concerns, and improvements. Uses low temperature (0.3) for focused, consistent feedback.

Tool: code_review

Parameter	Type	Required	Description
diff	string	Yes	The code diff or code to review
context	string	No	Focus areas or additional context (e.g. "focus on security")
model	string	No	Which model to use
maxTokens	number	No	Limit the response length. Default: 4096

Note: Use list_models to see available models, or pass any model name directly. Pro models aren't available on the free tier — if you see a quota error, switch to a Flash model or upgrade your plan.

---

Conversations

You can have multi-turn conversations where Gemini remembers what was said before. Pass the same conversationId across multiple generate_text calls:

1. First message: { "prompt": "What is photosynthesis?", "conversationId": "chat-1" }
2. Follow-up: { "prompt": "Explain it to a 5-year-old", "conversationId": "chat-1" }

Gemini will remember the first message when answering the follow-up.

Conversations expire after 30 minutes of inactivity.

---

License

MIT