Website Screenshot
Screenshot any list of web pages, full page or first screen, desktop or mobile: one row per URL with a direct image link, page title and HTTP status; lazy images are loaded and cookie banners hidden; pages that cannot be captured say why and are not billed. Not affiliated with the websites captured
Overview
Give it a list of web pages and get back one row per page with a link to a screenshot: the whole page from top to bottom or just the first screen, as a desktop browser or a phone shows it.
Screenshots answer questions text cannot: what a landing page, pricing page or ad actually looks like, whether a layout broke, what a competitor changed this week. They are also what a vision model needs when an AI agent has to "look" at a page. Before each screenshot the page is scrolled through so lazily loaded images appear, and common cookie-consent banners are hidden, so you see the page rather than an overlay.
Every URL gets a row with a clear status: captured, refused or verification page, gone, not loadable, or disallowed by the site. You never send 10 URLs and get 7 images back without knowing what happened to the other 3, and URLs without a screenshot are not charged.
Data notes
- This app opens the public pages at the addresses you give and captures how they look; the content shown belongs to the original websites, so follow their copyright and terms of use. This is not an official product of any of these websites and is not affiliated with them. The fee covers loading, rendering and storing the images, not the page content itself.
- Screenshot links work for 7 days after capture; download the images if you need to keep them longer. Anyone with a link can open it.
- Only the pages you list are captured; links on them are not followed, so the number of screenshots and the cost are known before a run.
- Full-page screenshots stop at 15,000 pixels; for longer pages the status details say so and give the full page height.
- Sites are treated politely: their robots.txt is respected, nothing on the page is clicked, and logins, paywalls, CAPTCHAs and interactive checks are never bypassed. Such pages come back with a status instead of a screenshot.
- Cookie banners are hidden, not accepted. Other popups (chat widgets, app download prompts, newsletter boxes) may still show; name them in "Elements to hide" to remove them.
- Not covered: pages behind a login, PDF and other files, and sites that block automated visitors outright (many social networks and large marketplaces).
What the results look like
Each row is one URL.
| URL | Status | Page title | Width (px) | Height (px) | Screenshot link |
|---|---|---|---|---|---|
| https://www.anthropic.com/news/claude-opus-4-5 | ok | Introducing Claude Opus 4.5 \ Anthropic | 1366 | 11529 | https://…/001-www.anthropic.com-57f01cae30.jpg |
| https://www.etsy.com | ok | Etsy - Shop for handmade, vintage, custom, and unique gifts for everyone | 780 | 1688 | https://…/001-www.etsy.com-bfc328c769.jpg |
| https://www.python.org/doesnotexist-xyz | not_found | Welcome to Python.org | 0 | 0 |
Each row also has the final URL after redirects, the HTTP status, device, image format, file size, full page height, link expiry time, status details and the capture time.
Use cases
- Let an AI agent see a page: pass "Screenshot link" to a vision model to check a layout, read a chart or compare designs.
- Monitor pages over time: capture competitors' home and pricing pages on a schedule and compare this week's images with last week's.
- Check the mobile view: capture the same pages as a phone to spot broken layouts before customers do.
- Keep evidence: archive how a landing page, an ad destination or a terms page looked on a given day, with the capture time on each row.
Scope and boundaries
Up to 30 URLs per run; only the given pages are captured, links are not followed. Full-page screenshots are cut at 15,000 px. Image links work for 7 days. robots.txt is respected; logins, CAPTCHAs and interactive checks are never bypassed, and nothing on the page is clicked. Billed per screenshot taken; URLs without a screenshot also get a row but are not billed. The fee covers loading, rendering and storing the image, not the page content
- Good for
- You need to see what a web page looks like, for yourself or to hand to a vision model: pass the screenshot link to the model
- You want to archive or monitor how pages look over time (home pages, pricing pages, competitor landing pages, ads landing pages)
- You need to check how a site looks on a phone: set device to mobile
- Do not use for
- You need the text of a page: use a web-page-to-Markdown app instead, which is cheaper to read than an image
- The page is behind a login or a paywall, or the site shows an interactive CAPTCHA: these come back as blocked
- Sites that refuse automated visitors outright (many social networks and large marketplaces) usually come back as blocked
- You need a PDF, a single element or a long-lived file: links expire after 7 days
Failure handling
Failure and retry behavior declared by the developer. We recommend including it in your system prompt when integrating.
- 1Every distinct URL gets exactly one row. `status` says what happened: ok, blocked, not_found, failed, robots_disallowed, invalid_url; only ok is billed (`billed_screenshots` = 1)
- 2blocked and failed can be retried later; status_text gives the reason. When a run runs out of time or is stopped, the URLs not yet captured come back as failed with that reason: run just those again
- 3If a popup still covers part of the page, run again with its CSS selector in hide_selectors
- 4A very heavy page (large news home pages) can need more memory than the app has: it comes back as failed and is not billed; capture it with full_page off or as mobile
- 5Download the image before link_expires_at if you need to keep it
Input
Parameters required to call this App, generated from the input.schema in manifest.json.
- URLs
urlsRequiredarray<string>Web pages to capture, 1–30. Each URL gets one row with a screenshot link or the reason there is none. A URL given twice is captured once.Itemsup to 30 itemsExample["https://www.anthropic.com/news/claude-opus-4-5"] - Full page
full_pageOptionalbooleanOn: capture the whole page from top to bottom (pages taller than 15,000 px are cut there, and the row says so). Off: capture only the first screen.Defaulttrue - Device
deviceOptionalstring(enum)desktop: a 1366 px wide desktop browser. mobile: a phone screen 390 px wide (images are 780 px wide at 2x), the way the site shows itself on phones.Allowed valuesDesktop (1366 px)desktopMobile (390 px)mobileDefaultdesktop - Image format
formatOptionalstring(enum)jpeg: smaller files, good for viewing and AI models. png: lossless, larger files.Allowed valuesJPEGjpegPNGpngDefaultjpeg - Extra wait (seconds)
wait_secondsOptionalnumberExtra time to wait after the page has loaded and settled, for slow animations or charts, 0 to 15.Range0–15Default2 - Elements to hide
hide_selectorsOptionalarray<string>CSS selectors of elements to hide before the screenshot, such as a chat widget or a promo bar. Common cookie-consent banners are hidden anyway.Itemsup to 20 itemsDefault
Output
Field structure of a single record, generated from the output.schema in manifest.json.
- URL
input_urlstringThe URL as given in the input, to match rows to inputsExamplehttps://www.anthropic.com/news/claude-opus-4-5 - Final URL
final_urlstringThe address actually captured; differs from the input when the page redirected. Empty when the page could not be openedExamplehttps://www.anthropic.com/news/claude-opus-4-5 - Status
statusstringok = screenshot taken; blocked = the site refused access or showed a verification page; not_found = the page does not exist (HTTP 404 / 410); failed = not captured (timeout, unknown domain, server error, file download, page too heavy for memory, or the run ran out of time; status_text says which); robots_disallowed = the site's robots.txt asks automated visitors not to open this page; invalid_url = the input is not a web address. Only ok is billedExampleok - Status details
status_textstringPlain-English explanation of the status with the reason when there is no screenshot; also says when a very tall page was cut at 15,000 pxExampleScreenshot taken
Sample data
Generated from output.schema. Fields without a sample value use structural placeholders.
{
"input_url": "https://www.anthropic.com/news/claude-opus-4-5",
"final_url": "https://www.anthropic.com/news/claude-opus-4-5",
"status": "ok",
"status_text": "Screenshot taken",
"billed_screenshots": 1,
"screenshot_url": "https://example.r2.cloudflarestorage.com/screenshots/20261113-c72dc30ff724df19/001-www.anthropic.com-57f01cae30.jpg?X-Amz-Expires=604800",
"link_expires_at": "2026-11-20T15:45:08+00:00",
"page_title": "Introducing Claude Opus 4.5 \\ Anthropic",
"http_status": 200,
"device": "desktop",
"format": "jpeg",
"width": 1366,
"height": 11529,
"file_bytes": 1226752,
"page_height": 11529,
"captured_at": "2026-11-13T15:44:50+00:00"
}Integration
This App can be integrated via MCP, API, SDK, or file export. All channels share the same capabilities and pricing. Every request authenticates with the Authorization: Bearer header using an API key (long-lived, created in the Data Hub Console); MCP clients can also sign in with OAuth, no key required. More options such as CLI and Skill are on the way. This App runs asynchronously and can be canceled midway: After you cancel, the run stops. Records already delivered are kept and billed.
With the MCP (Model Context Protocol), you can call this App directly from AI clients like Claude and Cursor. Pick your client and auth mode, then copy the config below.
Client config
Replace the value after Bearer with your long-lived API key. Works in any client, CI, or headless environment.
{
"mcpServers": {
"octoparse_datahub": {
"type": "http",
"url": "https://mcp-v2.octoparse.com?pin=MagicData/website-screenshot",
"headers": { "Authorization": "Bearer <YOUR_API_KEY>" }
}
}
}Let AI set it up for you
Don't want to edit configs by hand? Copy the install prompt and paste it into any AI client. It will complete the setup its own way. (By default it guides you through signing in, and never asks you to send your API key in the chat.)
detail.access.mcp.composeHint
Pricing
Metered by summing the output field billed_screenshots. Failed runs are not charged.
Multiple billing events accumulate independently. See each item for details. Failed runs are not charged.
View creditsTry it now
Fill in the parameters and run. Results come from a real call.