Cloudflare Browser Run on Containers Is Now Faster and More Scalable
Cloudflare has quietly delivered a major upgrade. The company rebuilt its Browser Run product on top of Cloudflare Containers, and the results are hard to ignore.
Developers can now spin up 60 browsers per minute through the Workers binding and run up to 120 concurrently. That is four times the previous limit. On top of that, Quick Action response times dropped by more than 50%. Existing users do not need to change a thing. The improvements are already live.
Cloudflare Browser Run on Containers represents a complete infrastructure overhaul, and one that pays off immediately.
Previously, Browser Run shared infrastructure with Browser Isolation (BISO), a separate Cloudflare security product. That arrangement created friction. BISO’s larger container images slowed startup times and made shipping new features harder. BISO’s long, steady user sessions also clashed with Browser Run’s short, spiky demand patterns. The outcome was scaling bottlenecks and availability delays for developers.
Moving to dedicated Cloudflare Containers resolved the foundational issue. However, it also introduced a new challenge: state management at scale.
As AI agent builders discovered Browser Run last year, request volumes surged quickly. Cloudflare had been relying on Workers KV to track container states. KV’s eventual consistency, around 30 seconds, became a critical bottleneck. A container could appear available in KV, but by the time a request routed to it, another request had already claimed it. That race condition led to browser overallocation and capped how fast Cloudflare could scale to meet demand.
The team therefore migrated container state management to D1, Cloudflare’s SQLite-based relational database. D1’s transactional model prevents two requests from claiming the same browser at once. To handle write volume, they batched updates using Cloudflare Queues, processing 100-row batches with a one-second timeout. The P95 batch write time now sits at 0.1 milliseconds. Capacity planning is, as a result, no longer a bottleneck.
Cloudflare also rethought how Quick Actions execute. Previously, the worker established a WebSocket to the remote browser and sent instructions one step at a time. Now, all parameters travel in a single HTTP request directly to the container, and the entire flow runs internally. The result is a sharp drop in average response time, less waiting, faster processing of DevTools Protocol messages.
Owning the container image independently means Cloudflare can also upgrade Chrome on its own schedule. Two long-requested features, WebGL for browser-based rendering and WebMCP (Model Context Protocol for the web), are now available. Both were blocked under the shared BISO infrastructure.
For developers building AI agents, this matters considerably. Cloudflare Browser Run on Containers is fast becoming a go-to platform for agentic web interaction. The company’s Agents SDK includes built-in Browser Run support, and a recently launched /crawl endpoint enables deep data extraction across entire websites.
Browser Run is available on all Workers plans. Developers can get started with the quick start guide or explore the Quick Actions API today.
