Summary: Cloudflare’s Browser Run Quick Actions expose ten common browser tasks — screenshots, PDFs, HTML, Markdown, link extraction, AI-driven structured data — as fixed REST endpoints or direct calls from a Worker. You skip the Puppeteer script entirely for these jobs. For small teams that just need a page rendered or scraped, that removes the whole headless-browser stack they would otherwise maintain.
What a Quick Action actually is
The ten endpoints are /content (HTML), /screenshot, /pdf, /markdown, /snapshot, /accessibilityTree, /scrape, /json, /links, and /crawl. Each is a fixed interface: you point it at a page and get the result back, with no page-object or selector plumbing unless the action needs it.
There are two ways to call them. Via REST, you create a custom API token with Browser Run - Edit permissions. Via a Workers binding, you add a browser binding to wrangler.json and call env.BROWSER.quickAction() directly — no API token needed:
{
"compatibility_date": "2026-03-24",
"browser": {
"binding": "BROWSER"
}
}
The compatibility date matters: .quickAction() requires 2026-03-24 or later, so older Worker configs will not see the method at all.
The /json action is the one with a real mechanism behind it. It uses Workers AI to return structured data shaped by a response_format.json_schema, which makes it closer to AI extraction than to plain scraping. Note that /crawl is the exception on access: REST only, not callable through the binding.
How you measure what it costs you
Every Quick Actions response carries an X-Browser-Ms-Used header reporting the browser time that request consumed, in milliseconds. The dashboard aggregates total Quick Actions requests and total browser hours. That combination means you can measure per-task cost yourself before committing a feature to it — sum the header values across a representative batch of pages and you have a real number, not a vendor’s.
The sources give no pricing or quota figures for browser hours, so you cannot compute a per-task cost in currency from the documentation alone. That gap is worth closing with your own measurement before scaling anything high-volume.
Where the simple tier ends
Cloudflare’s own docs draw the line explicitly: use Quick Actions for fast, simple common tasks, and use Puppeteer, Playwright, or CDP for advanced automation, custom workflows, or persistent browser sessions. That is the vendor saying these are canned endpoints, not a general browser runtime.
Local development is also degraded in two specific ways. Under npx wrangler dev, the /json action returns placeholder data because Workers AI is not available locally — fields with const, enum, or default in the schema get those values, everything else gets an empty value like "", 0, false, or []. And the Kitesurf browser is unavailable locally: passing browser: "kitesurf" throws an error. Setting "remote": true in the binding gets you real /json results and Kitesurf access during local dev. We have covered Kitesurf going open source under Apache 2.0 separately, so its local absence is a real gap for teams already using it.
Our take: Act now if you ship the boring jobs — preview images, PDF generation, Markdown feeds, link extraction — and you are a small team without platform engineers. Wiring one endpoint costs an afternoon, is reversible, and deletes a browser stack you would otherwise patch and scale. Before relying on it, measure X-Browser-Ms-Used across your real pages and check the dashboard total, because with no published pricing in the docs, the assumption that metered browser time stays cheap is just that, an assumption. Skip it if you already run Playwright or Puppeteer with persistent sessions and custom flows; Quick Actions cannot replace that, and migrating to fixed endpoints would be a downgrade. Two things would change the verdict: published browser-hour pricing that beats self-hosting at your volume, or /json extraction that disappoints on your actual pages — either pushes you back to a full browser framework.
The full Quick Actions documentation has the endpoint-level details and setup walkthrough if you decide to wire one in.
Sources
View the original post on X
