Skip to content
NLEN
Illustration: OpenAI pipeline found 948 Linux root bugs in one run: what autonomous bug hunting means

Cloudflare Browser Run: /snapshot now also returns PDF, links and JSON

By Ivo Donker · — text: Claude (AI)

What is new in /snapshot on Cloudflare Browser Run? Since November 19, the endpoint can return a PDF, the page's links, selected elements and AI-extracted JSON from a single page load. Before, /snapshot only returned HTML, screenshots, Markdown and the accessibility tree, so anyone who needed more had to load the same page a second time.

We read Cloudflare's changelog entry of November 19. The news spread through a post by @CloudflareDev on X and featured in the trending briefing of November 21.

What happened

Browser Run is Cloudflare's service for running a real browser in the cloud, often for agents that need to read web pages. Its /snapshot endpoint fetches several formats of the same page in one request. Four formats have been added: pdf, links, scrape and json. Each takes the same options as the single-format endpoint of the same name.

What happened
New formatWhat it returnsNote
pdfthe page as a PDF, with the same pdfOptions as /pdfprinted last
linksthe links on the page, optionally only visible or only internal onesoptions as in /links
scrapeelements matching the given selectorselements is required
jsonstructured data extracted from the HTML by Workers AIneeds prompt or response_format; incurs Workers AI usage

Details

Formats are captured in a fixed order rather than the order in the request. The PDF comes last because print handlers can change the page. A live page can change between captures, and if one format fails, the whole request fails. For JSON snapshots a single timeout covers everything, 60 seconds by default and 120 at most, including navigation, retries and the AI call. A snapshot still needs at least two distinct formats; for a single one, Cloudflare points to the matching endpoint.

What this means

For developers whose agents browse the web, this saves requests and time. An agent that wants to archive a page, follow its links and read a few fields used to need two or three separate rounds; now one will do. Because all formats come from the same page load, they also match each other better, with the caveat that a live page can change in between.

The JSON format is convenient but not free: it runs on Workers AI and counts toward that usage. And anyone fetching many pages also runs into the question of how publishers treat such traffic. Earlier we wrote about AI assistants in the browser, the open-source browser Kitesurf and a KV cache leak in Workers AI. The tension with publishers is the topic of our piece on robots.txt.

What we did not verify

We did not call the new formats ourselves or work out prices. What a JSON snapshot costs depends on the Workers AI model and is not stated in the changelog.

Checked on November 21, 2026.

This article was written with AI (Claude) and checked against the sources below.

Sources