A Guth Labs publication ↗
Back to News

Cloudflare adds four capture formats to Browser Run snapshots

Green glass prism among stacks of blank cream sheets cut into geometric shapes.
AI-generated editorial illustration

Cloudflare expanded its Browser Run snapshot endpoint on November 19 so one page load can produce PDFs, links, selected page elements and AI-extracted JSON alongside existing capture formats.

The official changelog says the endpoint previously returned HTML, screenshots, Markdown and accessibility trees. Workflows needing a PDF or structured extraction had to prepare the page again in another request.

JSON extraction uses Workers AI and incurs associated usage charges. Developers supply a prompt or response format; extracting selected elements requires specifying which elements to collect.

There are important limits. A request must include at least two distinct formats. Captures run in a fixed order, with the PDF last, and a live page can change between captures. Failure in any requested format fails the whole request.

For JSON snapshots, one timeout budget covers navigation, captures, retries and the AI request: 60 seconds by default, with a 120-second maximum.

The change reduces repeated page preparation for browsing and extraction workflows, but does not make every captured representation an identical, frozen view of the page.

Sources and citations

Revision history

Published .