Gemini MCP Server

Run a Gemini session. Similar to Codex but uses Google Gemini 3 Pro Preview.

Supports configuration parameters matching the Codex Config struct:

• prompt: The initial user prompt to start the conversation (required)

• cwd: Working directory context

• sandbox: Access policy ("read-only", "workspace-write", "danger-full-access")

• base-instructions: Override default system instructions

• developer-instructions: Additional developer context

• model: Optional override for model (default: gemini-3-pro-preview)

Continue a Gemini conversation by providing the conversation ID and prompt.

Use this to continue a multi-turn conversation started with the 'gemini' tool.

Generate images using Nano Banana (Gemini's native image generation).

Two models available:

• Nano Banana (default): Fast, cheap (~$0.04/image), good for most use cases

• Nano Banana Pro: Advanced model with better text rendering, infographics, diagrams

Auto-detection: Says "nano banana pro" or mentions text/infographic/diagram/chart/logo/poster in prompt → automatically uses Pro model.

Parameters:

• prompt: Text description of the image to generate (required)

• numberOfImages: How many images to generate (1-4, default: 1)

• aspectRatio: Image aspect ratio ("1:1", "3:4", "4:3", "9:16", "16:9", default: "1:1")

• usePro: Force Nano Banana Pro (auto-detected from prompt if not specified)

• outputPath: Optional path to save images

Generate a video using Veo 3.1 (Google's video generation model).

This starts an async video generation that takes 1-5 minutes. Returns an operation ID that you can use with gemini-video-check to poll for completion.

Parameters:

• prompt: Text description of the video to generate (required)

• aspectRatio: Video aspect ratio ("16:9" or "9:16", default: "16:9")

• resolution: Video resolution ("720p", default: "720p")

• firstFrameBase64: Optional base64 PNG image to use as first frame (from gemini-image)

Workflow:

1. Call gemini-video-generate → returns operationId

2. Wait 30-60 seconds

3. Call gemini-video-check with operationId → returns status or video

Check the status of a video generation operation.

If the video is still processing, returns the current status. If the video is complete, returns the video data.

Parameters:

• operationId: The operation ID from gemini-video-generate (optional - uses last operation if not provided)

• outputPath: Optional path to save the video file when complete