Scrape & research

scrape

Navigate the stealth (Camofox) browser to a URL and return the page's title and readable text. Use it for pages a plain HTTP fetch can't read — JS-rendered, logged-in, or bot-walled. Pass session (a saved logged-in identity) to read a page behind a login (e.g. your LinkedIn feed); without it the read is anonymous.

Returns { url, title, text }; with images: true it also returns { image, images } — the og:image (a profile display photo on a logged-in social profile) and profile-display-photo srcs the text extraction drops.

browser_export

Export the current tab to a downloadable artifact file (written on the Bureau host; returns its path — no base64). format:

• pdf · png (a live native screenshot of the viewport; fullPage for the whole page on CDP/Chrome drivers) · document (branded cover + clickable TOC with real page numbers + running header/footer) · html-zip (offline bundle) · html · text (markdown for agents, like llms.txt).

For document, set header / footer / cover / coverSubtitle / tocTitle. Pass path to choose the file (a relative path lands in the Bureau artifacts dir).

research_run / research_status

research_run starts a research report as a background job and returns a runId immediately. It distills the dataset at corpusPath, writes one chapter per facet, and renders a branded PDF — minutes of work. You author the reportConfig (chapters/facets); the run is deterministic. The chapter writer is swappable via model (default "claude-code" = the local CLI, no key; or any Anthropic API model id).

research_status polls a run by runId and returns { status: running | done | error }; on done, the report result — the corpusPath plus the render artifact (the PDF path on the Bureau host, pages, tocEntries).