Skip to content
Kitbay
MCP server

ai-vision-mcp

Runs locallyMITUpdated on Apr 6, 202680GitHub starsNeeds an API key

A local MCP server that sends images and videos to Google's Gemini models, through Google AI Studio or Vertex AI, and returns the analysis to the assistant.

What it does

A local MCP server that sends images and videos to Google's Gemini models, through Google AI Studio or Vertex AI, and returns the analysis to the assistant. Its tools describe an image, with modes for colour palettes, visual hierarchy and UI components, compare two to four images, detect objects and return an annotated image, audit a design by combining pixel measurements such as contrast with a model critique, and analyse YouTube links or local video files. It runs over stdio through npx and needs a Gemini API key or Vertex AI service account details. Setup is given for Claude Desktop, Cursor, Cline and Claude Code; the README suggests raising client timeouts above five minutes.

Capabilities

  • Describe an image or its UI components
  • Compare two to four images
  • Detect objects and return an annotated image
  • Audit a design's colours and contrast
  • Analyse YouTube or local videos