Vibeleaderboard
← All Intel
Intel / blog

The road to the agentic browser: A Kitesurf update

Source
blog.cloudflare.com
Author
Celso Martinho
Date

Context

Cloudflare's Kitesurf, a browser built for AI agents, can now expose a website's declared actions directly to an . Its update adds WebMCP support, letting an agent discover and call functions supplied by a site instead of reproducing the same action through screen clicks. The example is Cloudflare Radar, which lists tools such as navigate-to and set-location. This extends the earlier Kitesurf launch, which described a browser built around agent workloads. The benefit depends on the site exposing suitable tools: it gives agents an explicit interface for those actions.

Key takeaways · AI-distilled
  • Kitesurf now supports WebMCP, so an agent can call functions a site exposes, such as Cloudflare Radar's navigate-to and set-location tools, instead of simulating clicks on page elements.
  • Cloudflare reports that Kitesurf now passes more than 730,000 Web Platform Tests subtests, about 500,000 more than at its August launch, after adding CSS Layout, CSSOM, CSS Typed OM, custom elements, import maps and JSON modules.
  • To cut agent-loop latency, common DOM reads such as getAttribute and parentNode are now answered inside the Wasm DOM instead of crossing between the Boa JS engine and Wasm. Cloudflare says wall-clock time and CPU use stayed roughly at launch levels despite the added standards.
  • Kitesurf separates PageScript, the isolate that runs page code, from PageRenderer, which draws the pixels. Cloudflare used that split to render pages in a terminal through the Kitty graphics protocol or plain ANSI text, which also shows how an agent sees a page.
  • Kitesurf now covers the full Browser Run API, including CDP, Playwright, Puppeteer and , and is free during beta with per-account limits. Cloudflare says open-sourcing it is coming soon but gives no date.
Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • MCP — The Model Context Protocol — an open standard that lets any AI assistant plug into any tool or data source without custom integration code.
Read the source blog.cloudflare.com
Recommended reads
Comments

Checking sign-in…

Loading comments…