Screenshot API docs

URL in, image out. One endpoint, binary response, no JSON envelope to unwrap.

Pick your language

Screenshot API with curl — Take a screenshot of any web page from the command line. One request, binary image back, no SDK and no JSON envelope to unwrap.
Screenshot API for Node.js — Capture a web page as an image from Node. Uses the built-in fetch, so there is no dependency to install and nothing to keep updated.
Screenshot API for Python — Capture a web page as an image from Python. Standard library only - urllib, no requests, no SDK.
Screenshot API for PHP — Capture a web page as an image from PHP. Plain cURL, no Composer package required.
Screenshot API for Go — Capture a web page as an image from Go. Standard library only - net/http and net/url.
Web page to Markdown — the same render, read as text instead of photographed. Built for LLMs and agents.

What makes a capture correct

Most screenshot failures are not crashes, they are a 200 with the wrong picture. The things we handle, because a benchmark of 202 real pages caught each one:

Consent banners injected after load. A one-shot sweep loses that race; we install an observer before any page script runs.
Ad slots that leave a hole. Blocking the request does not reclaim the space, so the page keeps a grey gap and pushes content below the fold. We collapse them.
Pages that never go network-idle. Waiting for an event modern sites never emit turns a 3 second capture into a timeout. We wait for interactive, then spend one bounded settle budget.
Pages that wedge their own main thread. Every wait is bounded, and a page that stops responding is captured as it stands rather than held until a timeout.
Blank captures. Detected with two independent signals and reported, so a bot wall is never silently billed as a success.

Options

Parameter	Default	What it does
url	required	The page to capture. http/https only; private and reserved addresses are refused.
format	png	png, jpeg, webp or pdf.
full_page	false	Capture the whole scrollable page, scrolling first so lazy images load.
viewport_width	1280	Viewport width in pixels.
viewport_height	800	Viewport height in pixels.
selector	—	Capture only the element matching this CSS selector.
block_ads	false	Block ad networks and collapse the empty slot they leave behind.
block_cookie_banners	false	Remove consent banners, including ones injected after load.
color_scheme	light	Render the page as light or dark.
delay	0	Extra wait before capture, in ms.
cache	false	Serve a cached image when one exists. Cache hits are never billed.
fail_on_blank	false	Return 502 instead of an image when the capture looks blank.

Errors

Failures return JSON with a stable code. Successful captures return the image bytes directly, with no envelope.

{
  "error": {
    "code": "invalid_access_key",
    "message": "Unknown or revoked API key."
  }
}

Common codes: missing_access_key (401), invalid_access_key (403), quota_exceeded (402), render_timeout (504), blank_capture (502).

Failed renders are not billed, and neither are cache hits. We detect a blank capture rather than assuming a 200 means success, so a bot wall that returns an empty page does not come out of your quota.

Signed URLs

The headline use case is putting a capture straight into an <img src>. A raw key in public HTML is a billing incident waiting to happen, so sign the URL instead: the signature covers every parameter, and a tampered URL is refused.

node tools/sign.js "https://example.com" full_page=true

Set REQUIRE_SIGNATURE=true to refuse unsigned requests entirely.

Quota headers

Every response carries your position, so you never need a second request to check:

X-Quota-Limit: 2000
X-Quota-Used: 417
X-Quota-Remaining: 1583
X-Quota-Period: 2026-09