# Test notes: Kitesurf and Chromium

Date: 10 August 2026

## What I compared

This was a small, read-only comparison of Cloudflare Browser Run's Kitesurf beta and Chromium engines.

The comparison used nine public pages from Open for Agents, the Open for Agents demo and GTABoom. No form or tool was executed, and nothing on the tested websites was changed. The test used no login, customer data or model call.

## Results

| Measurement | Kitesurf | Chromium |
| --- | ---: | ---: |
| Successful pages | 9/9 | 9/9 |
| Screenshots | 9/9 | 9/9 |
| Markdown documents | 9/9 | 9/9 |
| Mean page load | 995 ms | 586 ms |
| Mean settled inspection | 4,133 ms | 2,419 ms |
| Mean first Markdown response | 2,569 ms | 1,162 ms |
| Reported Markdown browser time | 19,989 ms | 7,323 ms |
| Complete nine-page comparison | 42.7 s | 27.8 s |

Cached repeat extraction was approximately equal: 47 ms for Kitesurf and 45 ms for Chromium.

These results describe this nine-page comparison, not browsers in general. Each page was measured once, followed by an immediate repeat Markdown request to check cached consistency. The comparison did not measure engine CPU or memory use.

## WebMCP results

Chromium's experimental native surface exposed:

- 16 imperative tools;
- five declarative tools on Buddy, for 21 unique tools;
- six declarative tools on Kitchen, for 22 unique tools;
- no duplicate tool names;
- no legacy compatibility `unregisterTool()` or `provideContext()` methods; and
- no stale tools after navigation away.

The Kitesurf view available during this test did not provide the same native WebMCP interface. Code supplied by the site made the 16 imperative tools visible, but the five Buddy and six Kitchen declarative tools were absent from the tool list. The declarative forms remained present in the page markup. The site code also exposed older `navigator.modelContext` and `unregisterTool()` shapes.

This is a compatibility gap, not a security finding.

## Discovery documents

The public agent-actions, MCP draft and API Catalog documents all returned HTTP 200 with stable content. Those documents are separate from the tools attached to a page through WebMCP.

## Privacy and scope

- The comparison used public pages only.
- No form or tool was executed.
- No login, customer information or model call was used.
- Browser sessions were closed after the comparison.
- No screenshots, page bodies, cookies or visitor identifiers were retained.
- Nothing on the tested websites was changed.

## What this means

Open for Agents will continue using Chromium for browser testing and WebMCP inspection. Kitesurf remains worth revisiting as its WebMCP support develops or when a high-volume extraction workload makes its lower resource use especially valuable.

## Sources

- [Cloudflare Kitesurf announcement](https://blog.cloudflare.com/kitesurf/)
- [Cloudflare Browser Run](https://developers.cloudflare.com/browser-run/)
- [Browser Run WebMCP](https://developers.cloudflare.com/browser-run/features/webmcp/)
- [Browser Run pricing](https://developers.cloudflare.com/browser-run/pricing/)