Agents reviewing frontends, debugging layouts, or verifying deployed pages need to see them. The Screenshotty MCP server exposes the capture API as Model Context Protocol tools, so the agent calls take_screenshot like any native capability and the image flows back into its context.
The server is a small open-source Node package that talks to the Screenshotty API with your key — install it once per client and every session has screenshot capability.
claude mcp add screenshotty \--env SCREENSHOTTY_API_KEY=your_key_here \-- npx -y screenshotty-mcp
{"mcpServers": {"screenshotty": {"command": "npx","args": ["-y", "screenshotty-mcp"],"env": { "SCREENSHOTTY_API_KEY": "your_key_here" }}}}
URL, full-page, viewport, selector, format, and dark-mode options — the API surface, exposed as a tool schema.
Claude Code, Claude Desktop, Cursor, Windsurf, and any client speaking MCP over stdio.
Captures return as MCP image content, so vision-capable models see the page directly.
Runs locally with your API key in the environment — usage shows in your dashboard like any client.
take_screenshot (url or html, full_page, viewport, selector, format, dark mode, adblock/cookie-banner options) returning the image, and list_countries for geo-targeted captures.
The key is read from your local environment and sent only to api.screenshotty.link over HTTPS — the standard pattern for MCP servers wrapping an authenticated API.
Sign up at screenshotty.link (free tier: 100 screenshots/month, no credit card) and copy your key from the dashboard. Keep it in the integration's credential store rather than pasting it into shared workflows.
Typical loops: capture a local or deployed page after a change and review the layout, compare staging vs production visually, check a competitor's page, or generate assets via the html parameter — all without leaving the editor.