Screenshot

POST /screenshot

Some pages don’t survive being turned into text. A dashboard, a chart, a map, a pricing table whose meaning is in its layout — an agent needs to look at those. /screenshot renders the page in a real Chromium, waits for it to settle, and returns an image.

The response carries a short-lived image_url rather than the bytes, because that URL is what you hand to a vision model or drop into a report. Ask for include_image_data if you’d rather have base64 in the same response.

Tip

Over MCP the screenshot tool returns a native image block, so the model sees the page directly instead of being told about it.

Request

curl -X POST https://api.crisps.ai/screenshot \
  -H "Authorization: Bearer $CRISPS_API_KEY" \
  -H "content-type: application/json" \
  -d '{"url": "https://example.com", "full_page": true}'

Body parameters

Field	Type	Notes
url	string	Required. Must be a public http(s) address.
full_page	boolean	Capture the whole document instead of the viewport. Costs one extra credit.
width · height	integer	Viewport, default 1280×800. Clamped to 2560×2000.
device_scale	number	Pixel ratio, 1–2. Use 2 for retina-sharp output.
format	string	png (default) or jpeg.
quality	integer	1–100, jpeg only.
selector	string	Capture just this element instead of the page.
wait_for_selector	string	Wait for this to appear before capturing.
delay_ms	integer	Extra settle time after load, up to 5000.
dark_mode	boolean	Render with prefers-color-scheme: dark.
block_banners	boolean	Hide common cookie/consent overlays. Default true.
language	string	Browser locale, e.g. de-DE.
include_image_data	boolean	Also return the image base64-encoded in the body.

Response

{
  "url": "https://example.com",
  "final_url": "https://example.com/",
  "title": "Example Domain",
  "image_url": "https://api.crisps.ai/shots/N1c3-unguessable-name.png",
  "file": "N1c3-unguessable-name.png",
  "format": "png",
  "bytes": 48213,
  "width": 1280,
  "height": 800,
  "device_scale": 1.0,
  "full_page": true,
  "truncated": false,
  "expires_in": 86400
}

truncated is true when a full-page capture hit the height cap — very long pages are cut rather than refused. expires_in is how long image_url stays alive; captures are a cache, not storage, so fetch what you need and keep your own copy if it has to outlive the day.

Cost

2 credits per capture, 3 with full_page. A capture drives a real browser, so it prices above extraction.

What it will not fetch

Only public http(s) addresses. Anything resolving to a private, loopback or link-local address — or to a cloud metadata endpoint — is refused with url_not_allowed, including via a redirect. The same guard covers /extract, /map and /crawl.