The full upstream README, mirrored here for reference. Install config, tool schemas, adoption signals, and an original overview live on the MCP Screenshot Website Fast listing page.
Fast, efficient screenshot capture of web pages - optimized for CLI coding tools. Automatically tiles full pages into 1072x1072 chunks for optimal processing.
Built specifically for AI vision workflows, this tool captures high-quality screenshots with automatic resolution limiting and tiling for optimal processing by Claude Vision API and other AI models. It ensures screenshots are perfectly sized at 1072x1072 pixels (1.15 megapixels) for maximum compatibility.
Settings → Tools → AI Assistant → Model Context Protocol (MCP) → Add
Choose "As JSON" and paste:
Drop this into your client's mcp.json (e.g. .vscode/mcp.json, ~/.cursor/mcp.json, or .mcp.json for Claude).
Once installed in your IDE, the following tools are available:
take_screenshot - Captures a high-quality screenshot of a webpage
url (required): The HTTP/HTTPS URL to capturewidth (optional): Viewport width in pixels (max 1072, default: 1072)height (optional): Viewport height in pixels (max 1072, default: 1072)fullPage (optional): Capture full page screenshot with tiling (default: true)waitUntil (optional): Wait until event: load, domcontentloaded, networkidle0, networkidle2 (default: domcontentloaded)waitFor (optional): Additional wait time in millisecondsdirectory (optional): Directory to save screenshots - returns file paths instead of base64 imagescapture_selector - Captures a screenshot of a specific DOM element matched by a CSS selector
url (required): The HTTP/HTTPS URL to captureselector (required): CSS selector for the element to capturewidth (optional): Viewport width in pixels (max 1072, default: 1072)height (optional): Viewport height in pixels (max 1072, default: 1072)waitUntil (optional): Wait until event: load, domcontentloaded, networkidle0, networkidle2 (default: domcontentloaded)waitForMS (optional): Additional wait time in millisecondsselectorTimeoutMS (optional): How long to wait for the selector to appear before failing (default: 5000)Default usage (returns base64 images):
Save to directory (returns file paths):
Capture a specific element:
When using the directory parameter:
Captures a series of screenshots over time to create a screencast. Only captures the top tile (1072x1072) of the viewport.
url (required): The URL to captureduration (optional): Total duration in seconds (default: 10)interval (optional): Interval between screenshots in seconds (default: 2)jsEvaluate (optional): JavaScript code to execute at the startwaitUntil (optional): Wait strategy: 'load', 'domcontentloaded', 'networkidle0', 'networkidle2'waitForMS (optional): Additional wait time before startingdirectory (optional): Save as animated WebP to directory (captures every 1 second)Basic screencast (5 frames over 10 seconds):
Custom timing:
With JavaScript execution:
Save as animated WebP:
When using the directory parameter:
-w, --width <pixels> - Viewport width (max 1072, default: 1072)-h, --height <pixels> - Viewport height (max 1072, default: 1072)--no-full-page - Disable full page capture and tiling--wait-until <event> - Wait until event: load, domcontentloaded, networkidle0, networkidle2--wait-for <ms> - Additional wait time in milliseconds-o, --output <path> - Output file path (required for tiled output)The MCP server includes automatic restart capability by default for improved reliability:
For development/debugging without auto-restart:
Built specifically for AI vision workflows:
Contributions are welcome! Please:
PUPPETEER_SKIP_CHROMIUM_DOWNLOAD=true and provide custom executable--wait-for flag--wait-until strategiesMIT