Cloudflare Launches Kitesurf Agent-First Browser

Cloudflare's documentation reports that Kitesurf uses substantially less CPU and memory than Chromium for common agentic browser tasks, based on medians from five Browser Run executions on a Cloudflare test page.

• •Screenshot CPU: 380 ms for Kitesurf versus 1,173 ms for warm-pool Chromium, or 3.1x less CPU.
• •HTML-extraction CPU: 229 ms versus 877 ms, or 3.8x less CPU.
• •Screenshot memory: 57.8 MiB versus 271.0 MiB, or 4.7x less memory.
• •HTML-extraction memory: 39.4 MiB versus 273.7 MiB, or 7.0x less memory.

The same benchmark shows a latency cost. Cloudflare reports median wall times of 1,148 ms for Kitesurf screenshots and 820 ms for HTML extraction, compared with 637 ms and 472 ms for warm-pool Chromium, respectively. The company attributes Chromium's faster wall time in this comparison to a warm just-in-time compiler and Kitesurf's cold software renderer.

For teams operating large numbers of short-lived browsing sessions, CPU time and memory footprint often determine concurrency and infrastructure cost more directly than single-session latency. That industry trade-off is most relevant to scraping, retrieval pipelines, agent evaluation, and bursty task automation, where many jobs can run independently and a modest per-job delay may be acceptable.

Kitesurf exposes the Chrome DevTools Protocol (CDP), according to Cloudflare's playground and documentation. Existing clients that can drive Chrome through CDP can therefore connect to Kitesurf; Cloudflare documents activation by adding browser=kitesurf to Browser Run REST or CDP endpoints.

Cloudflare's public playground also provides example configuration for MCP clients using chrome-devtools-mcp, including Claude Code, OpenCode, Cursor, and Gemini CLI. This CDP compatibility matters operationally because it can reduce integration work for teams already using Playwright, Puppeteer, or MCP-based browser tooling, although workload-specific testing remains necessary.

The developer documentation says Kitesurf has passed more than 235,000 Web Platform Test subtests. It lists high coverage for several agent-relevant areas, including DOM, HTML, Selection, SVG, Encoding, CORS, and XHR. Those results indicate standards-test coverage, not universal compatibility with production websites.

Cloudflare also documents deliberate limitations: Kitesurf is slower in wall-clock terms than warm Chromium and does not aim for pixel-perfect rendering. The beta status and its stated rendering trade-offs make fallback behavior important for workflows that rely on visual fidelity, complex browser features, or latency-sensitive interactions.

Kitesurf's release extends the browser-automation options available to agent builders beyond a hosted or self-managed Chromium fleet. Its central technical distinction is using Workers-hosted V8 isolates and a browser implementation oriented around machine consumption of pages, rather than a full human-facing browser process.

Comparable agent systems often separate inexpensive extraction and navigation work from harder rendering or interaction cases. Cloudflare's published benchmark suggests Kitesurf could fit the lower-resource portion of that pattern, while Chromium remains a relevant comparison point where warm-start latency or rendering fidelity is the priority.