Sulat.com / Writing on AISaturday, November 21, 2026
Developer Tools/21 November 2026

Cloudflare’s Browser Run /snapshot now returns PDFs, links, scraped elements and extracted JSON

Browser Run's /snapshot endpoint can now return a PDF, a page's links, scraped elements and AI-extracted JSON from one page load, so a workflow no longer loads the same page twice.

Glossy 3D Cloudflare logo under a glass browser window, with PDF, link, JSON, screenshot and tag icons floating around it

At a glance

  • Browser Run's /snapshot endpoint can now return pdf, links, scrape and json alongside HTML, screenshots, Markdown and the accessibility tree.
  • Each new format reuses the options of its standalone endpoint. Scrape needs elements, and JSON needs a prompt or a response_format and uses Workers AI.
  • Request at least two distinct formats. They're captured in a fixed order with the PDF last, and one failed format fails the whole request.
  • JSON snapshots share one actionTimeout for navigation, captures, retries and the AI call: 60 seconds by default, 120 at most.
  • Quick Actions bill browser hours only: 10 minutes a day on Workers Free, 10 hours a month on Workers Paid, then $0.09 an hour.

Cloudflare’s Browser Run can now return a PDF, a page’s links, scraped elements and AI-extracted JSON from its /snapshot endpoint, all from one page load. The change is in a changelog entry dated 19 November. The Cloudflare Changelog account posted it on 20 November.

What changed

Until now, /snapshot gave you HTML, a screenshot, Markdown and the accessibility tree. If a workflow also wanted a PDF or structured data, it had to load the same page again in a second request. Four more values now work in the formats parameter: pdf, links, scrape and json. The post’s pitch is to skip the extra page loads and grab everything in one go.

Before and after: two page loads for eight formats, now one load returns all eight, four of them new
Four formats that needed a second page load now come back from the first.

What each new format takes

Each new format reuses the options of the standalone endpoint it matches.

  • PDF: the same pdfOptions as /pdf.
  • Links: visibleLinksOnly and excludeExternalLinks, as /links does.
  • Scrape: returns the elements you match with elements, in the same payload as /scrape. elements is required when you ask for it.
  • JSON: Workers AI pulls structured data out of the page HTML, using the prompt, response_format and custom_ai fields from /json. You need a prompt or a response_format, and the request uses Workers AI.
PDF, links, scrape and JSON cards, each with the options it reuses from its standalone endpoint
Each new format reuses the options of the endpoint it matches.

Cloudflare’s own example, from a Worker, asks for a PDF, the page links and the text of each paragraph:

Code
const response = await env.BROWSER.quickAction("snapshot", {
	url: "https://example.com/",
	formats: ["pdf", "links", "scrape"],
	elements: [{ selector: "p" }],
});

const { result } = await response.json();
console.log(result.pdf);
console.log(result.links);

How a request runs

You have to ask for at least two distinct formats. If you only want one, the single-format endpoint is still the route. Cloudflare captures the formats in a fixed order, not the order of your array, and prints the PDF last because print handlers can change the page. A live page can also change between captures. If any one format fails, the whole request fails.

For JSON snapshots, a single actionTimeout covers navigation, captures, retries and the AI request. It defaults to 60 seconds and tops out at 120.

Five snapshot rules: two formats minimum, fixed capture order, PDF last, one failure fails all, JSON clock
The PDF prints last, and a single failed format fails the whole request.

What it costs

Quick Actions are billed for browser hours only. The pricing page lists 10 minutes a day on Workers Free, and 10 hours a month on Workers Paid, then $0.09 per extra hour. A Quick Actions request that fails with a waitForTimeout error isn’t charged.

The JSON format adds a second meter, because it draws on Workers AI. The /json docs name Llama 3.3 70B Instruct as the default model. You can supply your own model and key through custom_ai instead.

How to call it

There are two routes. The REST API needs a custom API token with the Browser Run Edit permission. From a Worker, the quickAction method on the browser binding needs no token, but it wants a compatibility date of 24 March 2026 or later. Browser Run is the name Cloudflare gave Browser Rendering on 15 April.

In short

Browser Run’s /snapshot endpoint gained four formats on 19 November: pdf, links, scrape and json. They sit beside HTML, screenshot, Markdown and the accessibility tree, and one request can return any two or more of them.

The PDF prints last, one failed format fails the request, and JSON snapshots also use Workers AI.


Discover more from Sulat AI & Tech News

Subscribe to get the latest posts sent to your email.