Model Context Protocol — live at api.screenshotengine.com/mcp

Give Your AI Agent Eyes

Connect Claude Code, Claude Desktop, Cursor, or VS Code to the ScreenshotEngine MCP server. One tool call captures a clean, vision-ready screenshot of any website — ads and cookie banners already stripped.

Get Your Free API Key Read Documentation

Run once in your terminal

claude mcp add --transport http screenshotengine \
  https://api.screenshotengine.com/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Free tier includes 100 screenshots/month. No credit card required.

Vision for Your Agent in Three Steps

No SDK, no headless browser setup, no infrastructure. Your agent gets a take_screenshot tool and starts seeing the web.

Step 1

Get a free API key

Step 2

Paste one config

Add the MCP server to Claude Code, Claude Desktop, Cursor, or VS Code with the copy-paste snippets above.

Step 3

Ask your agent to look

“Screenshot my competitor's pricing page in dark mode” — the agent calls the tool and sees the result instantly.

Output Optimized for LLM Vision Models

A screenshot an agent can't interpret is wasted tokens. Every capture is tuned so vision models see the page, not the noise around it.

Ads, cookie banners, and popups stripped before capture — no overlay noise polluting the model's view
JPEG, PNG, and WebP with controllable viewport size to stay within vision-model image limits
Full-page capture (height: "full") for long documents and landing pages
CSS selector capture so agents inspect exactly the element they care about
Dark mode emulation and preset device viewports for responsive checks
Queue-less rendering in milliseconds — agent loops stay fast

Agent tool call

{
  "method": "tools/call",
  "params": {
    "name": "take_screenshot",
    "arguments": {
      "url": "https://example.com/pricing",
      "block_banners": true,
      "dark_mode": true,
      "height": "full",
      "format": "webp"
    }
  }
}

// → returns an image content block
//   the model can see directly

What Agents Do With Eyes

Once your agent can see rendered pages, whole categories of tasks stop needing a human in the loop.

Frontend verification

Coding agents screenshot the page they just changed and confirm the fix actually rendered — before telling you it works.

Competitor & market research

Research agents capture live pricing pages, landing pages, and SERPs as clean images instead of parsing brittle HTML.

Monitoring & QA loops

Scheduled agents watch key pages for visual regressions, broken layouts, or unexpected content changes.

Archival & reporting

Agents export pages to PDF for compliance snapshots, invoices, and reports — straight from the same tool.

The `take_screenshot` Tool

One tool, every capture mode. Images come back inline so the model sees them immediately; PDFs arrive as embedded resources.

Parameter	Description
url	The URL to capture (required)
output	"image" (default) or "pdf"
format	"jpeg" (default), "png", or "webp"
width / height	Viewport size in pixels; height "full" captures the entire page
block_banners	Strip ads, cookie banners, and popups before capture
dark_mode	Emulate the dark color scheme
viewport_device	Preset device viewport, e.g. iphone-15, macbook-pro-16
selector	CSS selector to capture a single element
wait_for	Milliseconds to wait after page load before capturing

Machine-readable server card at /.well-known/mcp/server-card.json

Full REST API with OpenAPI spec at /openapi.json

LLM-friendly site summary at /llms.txt

MCP Server FAQ

What is the ScreenshotEngine MCP server?

It is a hosted Model Context Protocol (MCP) server at https://api.screenshotengine.com/mcp that exposes a take_screenshot tool. Any MCP-compatible AI agent — Claude Code, Claude Desktop, Cursor, VS Code, and others — can call it to capture a live screenshot of any public URL and see the result directly.

How do I connect Claude Code or Cursor to ScreenshotEngine?

For Claude Code, run: claude mcp add --transport http screenshotengine https://api.screenshotengine.com/mcp --header "Authorization: Bearer YOUR_API_KEY". For Cursor, add the server URL and Authorization header to ~/.cursor/mcp.json. A free API key takes under 60 seconds to create — no credit card required.

Do MCP protocol calls count against my screenshot quota?

No. Handshake and discovery calls (initialize, tools/list, ping) are free. Only successful screenshot captures count toward your monthly limit — the same Pay Only for Success policy as the REST API.

Why are ScreenshotEngine captures better for LLM vision models?

Ads, cookie consent banners, and popups are stripped before capture, so the model sees the actual page content instead of overlay noise. Output size and format (JPEG, PNG, WebP) are controllable to stay within vision-model image limits, and full-page or CSS-element capture lets agents look at exactly the region they care about.

What outputs does the MCP take_screenshot tool support?

Images (JPEG, PNG, WebP) are returned inline so the model can see them immediately, and PDFs are returned as embedded resources. Scrolling video capture (WebM) is available through the REST API, since video payloads are too large for MCP responses.

Your Agent Is Flying Blind

Fix that in the next 60 seconds — free key, one config line, and it can see any page on the web.

Get Your Free API Key

No credit card required for the free tier.