Screenshots of Pages
Part of the MountDev Ops Websites documentation. Updated November 21, 2026
MountDev Ops Websites documentation | Updated November 21, 2026
A screenshot is a real picture of a page of your website, drawn by a real browser, for your own AI assistant. It works for a page in its draft state (the working copy, with changes waiting for review) or as published, at the width and color the assistant asks for, with no publishing. It exists so an assistant that builds a page can see what it built before it tells you the page is ready.
What an Assistant Gets
One call returns:
- Drew: one plain sentence saying the website, the page, whether it is the working copy (draft) or the published version, how many changes waiting for review it included, the views drawn, and what is not drawn. An assistant never mistakes the draft for what is live.
- Pictures: JPEG slices at most 1,500 pixels tall, in order, each with a caption naming the width, the color, the version, and the part of the page. A call holds at most 12 pictures and 6 MB. A whole page or chosen sections (up to 12, by place) can be asked for.
- Structure (only when asked): a plain text outline of the page, its header, menu, sections with their height, and footer, up to 20,000 characters.
- Checks (only when asked): sideways scrolling, pictures that did not load, empty sections, text that is hard to read, links and menu items to pages that do not exist, errors reported while drawing, small tap targets at phone widths, cut off text, and slow loading. Each check has a fixed name and one plain sentence, and a check that finds nothing says so. The checks that need only data work even when the browser is not available.
Widths and Colors
A width is a whole number from 320 to 1920 pixels, or the words phone (390), tablet (820), or desktop (1440, the start). Up to 3 widths and 6 views (a width and a color each) in one call. Light is the start. Dark is drawn only when the website has dark rules in its theme stylesheet or a section; otherwise the answer says the website has no dark version and draws light.
What Is Not Drawn
As in Preview: code snippets, scripts in a Custom HTML section, the chat bubble, forms (a placeholder is drawn), and published tools (a card is drawn). The answer lists any of these the page uses, so they are not reported as missing.
How It Runs
The action sites.page.screenshot (tool websites_page_screenshot) checks the input and the rates, adds the render to your account's own queue, and waits up to 45 seconds for it. If it finishes, the call answers with everything. If not, it answers running with a screenshot_id, and sites.screenshot.get (tool websites_screenshot_result) answers the same way later: the whole answer, still running, or why it failed. It works for 30 minutes after the render finished and only for the connection that asked.
The browser is never given a real address of your website. It opens a made-up address, and your workspace answers every request the page makes from your own draft or your own stored published version. Anything on another website loads the way it does for a visitor, only if it is a public https address. No form is sent and nothing of your account goes to another website.
What Is Kept
The pictures and text of an answer are kept in your account's own database for 30 minutes after the render finished, then removed. What stays is one row of counts: the checks by level, the number and size of the pictures, the browser time, and the requests. No picture, page text, or address of another website is in a log. The Activity Log has one line for each render, made by the AI agent or the key, with the website, the page, the version, the widths and colors, the counts, the pictures, the bytes, and the browser seconds. Browser time is counted in your account's own usage and is not billed.
Limits
3 renders a minute and 60 an hour for one connection; 3 at once and 300 a day for your account; one at a time for each website; a page of 2 MB of HTML or less; 90 seconds of browser for one call. A refusal says when to try again.
The Agent Guide and the Ready Job
The Agent Guide page for Websites (mountdev://guide/app/sites) has a section, Seeing a Draft Page Before It Is Published. It tells an assistant when to take a screenshot (after saving a page and before saying it is ready, after changing a theme, menu, header, or footer, and always at the phone and desktop widths), the steps in order (save, screenshot with checks, look, fix, draw again, say plainly what was found and fixed), what a screenshot cannot see, the limits, and what the answers running and failed mean. It also says never to call a page fine without a screenshot and never to call a draft live.
A ready job, Check a Page Before You Hand It Over (Websites), gives an assistant those steps for one website and page. It is listed only for an account that has Websites. A person starts it by asking their assistant to check a page.
Who Can Use It
Anyone who can use Websites, through an assistant connection or a key of the account. The queueing tool is a Write permission, because it adds a row to your queue, and the result tool is Read. A render never publishes and never changes a page, a menu, or a setting. Export Everything includes the counts row (sheet Website Screenshots) and not the held pictures.
