Labsco
MCP SERVER

Gemini MCP

by houtini-ai ↗

Gemini's image, video, SVG and search-grounded answers as tools inside the assistant you already use.

Model Routing, Multi-Model Consultation & Cost Control
Summary
Grounded search, images, SVG and video — the four things worth reaching to another model for.

Search grounding is the quiet one: answers come from a Google search run at question time rather than from a training cutoff, with the sources attached. SVG output is the underrated one, because it returns editable vector code instead of a raster you have to redo. Two limits to plan around: deep research takes minutes, so keep max_iterations to 3 or 4 under Claude Desktop's 4-minute tool timeout and push it higher in an IDE; and media resolution is a real cost dial, with MEDIA_RESOLUTION_MEDIUM giving the same OCR quality as high at half the tokens.

What it is

A TypeScript server that puts Google's Gemini models behind MCP tools — chat with Google Search grounding, image and video generation, SVG output, and image analysis — with results previewing inline rather than landing as file paths.

What you get
  • gemini_chat — answers with Google Search grounding on by default, sources returned as markdown links, and a thinking_level you can raise or lower
  • gemini_deep_research — repeated grounded searches synthesised into a report, steered with focus_areas and bounded by max_iterations
  • generate_image — image generation with optional search grounding, so a chart can be drawn from current data; full resolution saved to disk
  • edit_image — natural-language edits with multi-turn continuity through thought signatures
  • generate_svg — real SVG code in technical, artistic, minimal or data-viz styles, ready to commit or embed
  • generate_video — Veo 3.1 video with native synchronised audio, aspect ratio, seed and up to 4 variations per call
  • describe_image and analyze_image — a fast description, or structured extraction and detailed reasoning
  • load_image_from_path — a local file read in as base64 for any of the image tools
  • generate_landing_page — a self-contained HTML page with inline CSS and vanilla JS
  • gemini_prompt_assistant, gemini_list_models and gemini_help — prompting guidance, the live model list, and documentation without leaving the client
Requirements

A Gemini API key from Google AI Studio in GEMINI_API_KEY — the free tier covers most development use, with rate limits showing up first on deep research. Node.js 18 or higher; run with npx from @houtini/gemini-mcp. Optional: GEMINI_IMAGE_OUTPUT_DIR to auto-save generated images and video, and ffmpeg on PATH for video thumbnails.

Setup effort

One command plus a key — npx -y @houtini/gemini-mcp, then supply credentials