Google Gemini MCP
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| DEBUG_MCP | No | Log to stderr for debugging tool calls | false |
| GEMINI_API_KEY | Yes | Google AI API key from Google AI Studio | |
| GEMINI_MCP_LOG_FILE | No | Write logs to ~/.gemini-mcp/logs/ | false |
| GEMINI_DEFAULT_MODEL | No | Default model for gemini_chat and analyze_image | gemini-3.1-pro-preview |
| GEMINI_IMAGE_OUTPUT_DIR | No | Auto-save directory for generated images | |
| GEMINI_DEFAULT_GROUNDING | No | Enable Google Search grounding by default | true |
| GEMINI_ALLOW_EXPERIMENTAL | No | Include experimental/preview models in auto-discovery | false |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {
"listChanged": true
} |
| resources | {
"listChanged": true
} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| gemini_helpB | Get comprehensive help about Gemini MCP features, settings, and best practices |
| gemini_prompt_assistantA | Get expert prompt templates and guidance for Gemini image generation. Covers photography (portraits, products, cinematic), chart/diagram design (9 professional design systems including FT, Bloomberg, Tufte, Du Bois), lighting, colour grading, lens simulation, and style aesthetics. For charts: use chart_design with a color_scheme to get a full professional design system prompt. |
| gemini_chatA | Chat with Google Gemini models. Grounded in Google Search by default, on gemini-3.1-pro-preview. DO NOT SET max_tokens - the server allocates the model's full output ceiling automatically. It is a cap, not consumption, so unused headroom costs nothing; setting a small one makes Gemini 3 thinking burn the whole budget and return empty output that looks like a timeout. [MCP_RECOMMENDED_TIMEOUT_MS: 300000] |
| gemini_list_modelsA | List available Gemini models and their descriptions |
| gemini_deep_researchA | Conduct deep research on complex topics using iterative multi-step analysis with Gemini. This performs multiple searches and synthesizes comprehensive research reports (takes several minutes). [MCP_RECOMMENDED_TIMEOUT_MS: 900000] |
| describe_imageA | Analyze and describe one or more images using Google Gemini image models (Nano Banana Pro). Returns a text description — no image is generated. Default model: gemini-3-flash-preview. [MCP_RECOMMENDED_TIMEOUT_MS: 180000] |
| analyze_imageA | Analyze and extract information from one or more images using Gemini multimodal understanding. Returns a text analysis - no image is generated. Default model: gemini-3-pro-preview. DO NOT SET max_tokens - the server allocates the model's full output ceiling automatically; a small cap is spent on Gemini 3 thinking and returns empty output that looks like a timeout. [MCP_RECOMMENDED_TIMEOUT_MS: 300000] |
| load_image_from_pathA | Read a local image file and return it as base64-encoded data ready to pass to generate_image, edit_image, describe_image, or analyze_image tools. Supports JPEG, PNG, GIF, WebP, BMP. |
| generate_imageA | Generate an image using Google Gemini image models (Nano Banana Pro). Returns image with inline preview in Claude Desktop and saves full-resolution to disk. Default model: gemini-3-pro-image-preview. |
| edit_imageA | Edit one or more images using Google Gemini image models (Nano Banana Pro). Provide images and natural-language instructions for how to modify them. Returns edited image with inline preview and saves full-resolution to disk. |
| generate_landing_pageA | Generate a complete, self-contained HTML landing page using Gemini. Returns inline preview with responsive viewport controls. No external dependencies; inline CSS and vanilla JS only. [MCP_RECOMMENDED_TIMEOUT_MS: 300000] |
| generate_svgA | Generate scalable vector graphics (SVG) using Gemini. Creates clean, production-ready SVG code for diagrams, illustrations, icons, and data visualizations. Returns inline preview with SVG viewer. [MCP_RECOMMENDED_TIMEOUT_MS: 240000] |
| generate_videoA | Generate videos using Google Veo 3.1 AI model. Creates realistic 4-8 second videos from text prompts with optional first-frame image and reference images for character/style consistency. Supports native audio generation. Processing time: 2-5 minutes for 1080p videos. Returns video file path with optional thumbnail and HTML preview player. ⚠️ IMPORTANT: Video generation is ASYNC and takes 2-5 minutes. The tool will poll for completion automatically. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
| gemini-image-viewer | |
| gemini-video-viewer | |
| gemini-svg-viewer | |
| gemini-landing-page-viewer |
TDQS
Scored across 13 tools
Most tools are clearly distinct (help, list_models, chat, research, image generation/editing, video generation). However, describe_image and analyze_image overlap significantly—both analyze images and return text—with only subtle differences in default model and phrasing, which could cause misselection. Also, gemini_help overlaps with what an agent might expect from general documentation but is distinct enough.
Most tools follow a verb_noun pattern (gemini_list_models, generate_image, edit_image, generate_landing_page, generate_svg, generate_video), but some tools omit the 'gemini_' prefix (describe_image, analyze_image, load_image_from_path) creating minor inconsistency. The use of 'generate' for different output types is clear, but 'describe' vs 'analyze' could be more distinct. Patterns are mostly predictable.
With 13 tools, this server is well-scoped for a multimodal AI assistant covering chat, research, image, video, and text generation. Each tool has a clear purpose and covers distinct capabilities (help, models, prompting, chat, deep research, image in/out, editing, landing page, SVG, video). The count is appropriate without being excessive.
The tool surface covers the key workflows: image generation (generate_image), image editing (edit_image), image analysis (describe/analyze_image), local image loading (load_image_from_path), video generation (generate_video), text generation (generate_svg/landing_page), and interactive use (chat, deep_research). A clear lifecycle exists for image tasks (load→analyze→generate/edit). Missing features like image manipulation beyond editing or direct video editing are minor and likely out of scope.