Page Scanner

Sets up Page Scanner's helper, which Chrome starts to connect the extension to this server, in every Chromium browser on this computer. Needs no port and no token, and returns no secret. Only needed once per machine. Afterwards the user opens the Page Scanner settings in Chrome, Local Agents, presses Connect and allows the prompt; tell them that, since you cannot do it for them.

The older way to connect, for a machine where the install tool cannot set up the helper: writes the port and token the Page Scanner extension needs, and returns them so you can tell the user what to paste. Prefer install. The token is a secret that lets anything holding it scan any tab the user has open, and it will appear in this conversation, so do not repeat it anywhere else.

The Chrome profiles that have Page Scanner paired and connected right now. Each is a separate browser: use its browserId to say which one you mean.

Takes no parameters.

The windows and open tabs of one browser, so you can pick what to scan. Tabs carry the windowId they belong to, and windows say which one is focused.

Captures a whole web page and writes it to disk. Name a tabId from list_tabs to capture a tab that is already open, a url to open one and capture that, or urls to capture a list of pages one after another. A PDF keeps real, selectable text; PNG and JPEG are images. Returns the absolute path it wrote, and a truncated report when the page was too large to capture whole; for urls, one result per page, each with its path or its error. Set markdown to get the page as text too, which is far easier to read than a PDF. A tall page or a long list can take minutes. When the answer has status "running", the capture is still going in the browser: call this tool again with exactly the same arguments to wait for its result, and do not start another capture of that tab meanwhile.

Reads a page's design as it is drawn: the colors, type, spacing, radii, shadows, motion and breakpoints its elements use, and the custom properties it declares, in both color schemes. Writes tokens.json (W3C design tokens), tokens.css, tailwind.preset.js, audit.md (values nearly equal), contrast.md (WCAG contrast of every text color over its background) and extract.json (the raw measurements) into outputPath, a directory. Give urls for several pages of one site, or crawl with a start address to have the pages found by following its links. A page that answered an HTTP error is not read in, and in a crawl a page that a redirect took to another site or to a page already read is left out (pages[].left says why). Inferred tokens are numbered by use, not named; name them yourself if the user wants names. A tall page or a long list can take minutes. When the answer has status "running", the capture is still going in the browser: call this tool again with exactly the same arguments to wait for its result, and do not start another capture of that tab meanwhile.

Scrolls a page from its top at a steady pace, 1.25 screens a second, and records how it behaves: where the screen holds while its content plays, what fades, slides, scales or is uncovered, and whether it glides on or plays once when the scroll stops. The scroll is cut into scenes. Writes a PDF (an overview and a page per scene with six frames from its entry to its exit), a .json beside it with every scene measured, and each scene's six frames as JPEGs in a .frames folder. Use it to rebuild a page's scroll animations, or to show them without a video. The result is that JSON, every frame with its absolute path, and the files' paths. The scan takes up to two minutes; a tabId is scrolled where the user sees it, and their touching the page or the popup's Stop ends it early with what it had (ended says so). A tall page or a long list can take minutes. When the answer has status "running", the capture is still going in the browser: call this tool again with exactly the same arguments to wait for its result, and do not start another capture of that tab meanwhile.

Compares two captures of the same page that scan_page wrote, with no browser needed. Two Markdown files are compared a passage at a time and the answer carries the changed lines and a unified diff; two PNGs are compared pixel by pixel and the newer one is written again with the changed regions outlined. For PDFs, scan with markdown "beside" and pass the PDFs or their .md files. "section" limits a text comparison to one section by its heading. "changed" says whether anything did.

Before quoting a page in an answer, checks that each quote or value occurs in the capture’s Markdown as written, with no browser and no model. The Markdown’s own markup (escapes, emphasis, link syntax, list and quote prefixes) is undone and line wrapping ignored; a quote never runs across two blocks, and a word or number has to stand apart (4.99 is not in 14.99). Each quote comes back with the line and the section it is under, or, when missing, where the capture parts from it, to correct it by. A match only once curly quotes, dashes and case are folded is reported as "loose" and does not count as found unless loose is set. "allFound" says whether every quote is there.