MCP Screenshot

What it does

Integrates screenshot capabilities into AI agents, allowing them to capture full screen, specific regions, or web pages and analyze them using Google Vertex AI/Gemini models. It includes a history system with BM25 text search and perceptual hashing for finding similar images.

Tools

• capture_screenshot: Captures screens or URLs with support for zooming and regions.
• describe_image: Uses Gemini AI to generate detailed descriptions of images.
• verify_visualization: Analyzes charts and graphs using specialized expert modes (d3, chart, graph).
• search_screenshots: Search historical captures using text queries.
• find_similar_images: Locates visually similar screenshots using perceptual hashing.
• combined_search: Hybrid search using both text and image similarity weights.

Installation

{
  "mcpServers": {
    "screenshot": {
      "command": "python",
      "args": ["-m", "mcp_screenshot.mcp.server"],
      "env": {
        "VERTEX_AI_PROJECT": "your-project-id",
        "VERTEX_AI_LOCATION": "us-central1"
      }
    }
  }
}

Supported hosts

Confirmed support for Claude Code and general MCP clients.