Browser Run

At a glance

Browser Run is worth evaluating when your application needs a real browser but you do not want to maintain a Chrome fleet. Cloudflare runs the browser on its infrastructure: Quick Actions return a screenshot, PDF or extracted content from one request, while Browser Sessions let your code control a stateful browser. The useful choice is how much control the job needs. A screenshot does not require building a full browsing agent.

Good fit

Developers rendering JavaScript pages, capturing a known view or automating a sequence with Puppeteer, Playwright or the Chrome DevTools Protocol. Use a Quick Action for one output and a session when later actions depend on earlier state. Existing Cloudflare Workers can use a browser binding; clients elsewhere can use documented remote interfaces. TinyFish is another candidate when the desired interface is a natural-language task, though its suite also includes directly controlled browsers.

Consider something else when

You need automation that every destination will accept, or expect a cloud browser to behave exactly like a signed-in local browser. Cloudflare adds identification headers that cannot be removed, and destination sites can block automated traffic. Stateful sessions also need lifecycle management: the documented default inactivity timeout is 60 seconds. Reusing a browser does not by itself isolate users; separate contexts are needed for separate cookies and caches.

Start here

Start with the documented screenshot Quick Action using a public page you control or have permission to capture. A REST request needs your Cloudflare account ID and a token with Browser Run - Edit permission; a Worker can use its binding instead. Check that the returned image contains the intended content and record the X-Browser-Ms-Used response header. Dynamic pages may need an explicit wait condition. For a later session trial, close the browser in cleanup rather than leaving it billed until timeout. These are proposed checks; no account, token or Browser Run request was used here.

Cost and requirements

On 31 October 2026 the Workers Free allowance was 10 browser minutes per day and three concurrent sessions. Workers Paid included 10 browser hours per month, then $0.09 per additional hour. Sessions also included ten concurrent browsers averaged monthly, then $2 per additional browser under the documented daily-peak calculation. Quick Actions have browser-time charges without the session-concurrency charge. These are incremental browser rates, not the whole Workers bill. AI extraction through /json also consumes Workers AI or a separately chosen model service.

Limits to know

Cloudflare’s FAQ describes ordinary rendering as ephemeral, but Quick Action output is cached for five seconds by default, configurable up to a day or disabled with cacheTTL set to zero. Asynchronous crawl results are retained for 14 days; opt-in session recordings for 30 days. Do not treat all modes as zero retention. Beta outbound domain restrictions can narrow session traffic, but omitting both allowlist fields permits unrestricted HTTP/HTTPS destinations. Remote processing, cookies and supplied headers need a data decision. This assessment used primary documentation; rendering fidelity, throughput, isolation and service reliability were not tested.

Related work

• TinyFish: natural-language web tasks alongside search, fetch and browser APIs (alternative)

Still unclear

• Rendering accuracy and waiting behavior on the intended pages remain untested.
• Actual browser-time, concurrency costs and destination acceptance need a representative workload.