Website Screenshot and PDF API

Give it URLs, get back a PNG, JPEG or PDF of each page, plus one structured row per URL saying what happened. Headless Chromium, no API key of your own, no browser to run, no machine-learning models.

Built to be called by scripts and AI agents: the input is a flat JSON object, the output is one dataset row per URL in the order you sent them, and every failure says why in a sentence.

Input

Field	Type	Default	What it does
urls	array	-	Public http(s) pages. Up to 100 per run.
format	png / jpeg / pdf	png	Image or print-layout PDF.
device	desktop / mobile / tablet	desktop	1280x800 at 1x, 390x844 at 2x with touch, 820x1180 at 2x with touch.
viewportWidth, viewportHeight	integer	preset	Override the preset's size (200-3840 by 200-4000).
fullPage	boolean	false	Capture the whole scrollable page. Scrolls once first so lazy images load. Images only.
maxFullPageHeight	integer	12000	CSS pixels. Taller pages are cut here and the row says truncated: true. Capped at 8192 for the 2x presets, which is Chromium's texture limit.
waitUntil	load / domcontentloaded / networkidle	load	When navigation counts as done.
delayMs	integer	0	Extra wait before capturing, up to 30000.
waitForSelector	string	-	CSS selector that must be visible before capturing. Not found in time means the URL fails, uncharged.
darkMode	boolean	false	Sets prefers-color-scheme: dark.
timeoutSecs	integer	30	Per-page timeout, 5-120. If the load event is not reached in time but the document is there, the page is captured as it stands and the row says wait_timed_out: true. A page that never responds fails.
captureErrorPages	boolean	false	Capture pages that answer 4xx/5xx instead of failing them.
pdfFormat, pdfLandscape, pdfPrintBackground		A4, false, true	Paper size, orientation and whether backgrounds print.

Output

The file goes to the run's key-value store and each dataset row carries its key and public URL. One row per input URL, always, in input order:

index, url              position and the URL you sent
status                  ok | failed | blocked | skipped
error                   why it is not ok, in one sentence; null when ok
final_url, title        where the browser ended up, and the page title
http_status             status of the main document
output_key, output_url  the file in the key-value store, and its public URL
content_type, output_bytes
width, height           pixels of the delivered image (null for PDF)
pages                   page count of a PDF (null for images)
truncated               a full-page capture was cut at maxFullPageHeight
wait_timed_out          the wait condition was not met in time; captured as it stood
format, device, viewport_width, viewport_height, full_page
blocked_requests        sub-requests the page made at private addresses that were refused
billed_event            what this row was charged under; null if not charged
navigation_ms, capture_ms, total_ms

Billing

Pay per event: a small actor-start fee, then one charge per file produced.

• screenshot-captured for a PNG or JPEG, pdf-rendered for a PDF.
• Only a file that was produced, stored and is under the size limit is charged. A URL that fails, times out, answers 404, is refused as private, or whose selector never appears gets a row with the reason and costs nothing.
• The run stops as soon as your spending limit is reached. URLs not reached get a skipped row so nothing silently goes missing.

What it will not fetch

This Actor renders pages for anyone who can start it, so it refuses to be a way into someone else's network. It only fetches public http and https URLs:

• file:, chrome:, ftp:, data: and every other scheme are refused.
• Loopback (localhost, 127.0.0.0/8, ::1), private ranges, link-local including the cloud metadata address 169.254.169.254, CGNAT, multicast and their IPv6 equivalents are refused - in every spelling a browser accepts, including decimal, hex and octal forms and IPv4-mapped IPv6.
• A name that resolves to any private address is refused.
• The check repeats for every redirect and for every request the page itself makes, so a public page cannot make the browser fetch an internal address, and the browser's connections go through a proxy that connects only to the address it checked, which defeats DNS rebinding.
• Ports used by mail, databases, SSH and similar services are refused.

A refused URL gets status: blocked.

Limits

100 URLs per run, 25 MB per file, full-page height capped as above, one page timeout per URL. Pages that need a login, solve a CAPTCHA or sit behind a bot wall are captured as the wall, not bypassed.