Cloudflare added WebMCP support to Kitesurf, its browser for AI agents. On sites that expose compatible tools, an agent can invoke a function directly instead of navigating only by simulated clicks. Cloudflare is showing the feature in a public playground. The update expands a way to interact with supported sites; it does not establish that Kitesurf can complete every browser task or that all sites expose useful tools.
Cloudflare added WebMCP support to Kitesurf on September 28. Kitesurf is its browser for AI agents running on Cloudflare Workers; the browser itself was introduced in August. WebMCP lets a compatible site expose named actions that an agent can invoke directly. An agent can therefore call a site’s search or navigation function when it is offered, instead of finding a visual control and simulating a click for every step.[1], [2]
Cloudflare demonstrates the update in a public Kitesurf playground. On a Cloudflare Radar page, the developer-tools panel shows exposed actions such as navigation and setting a location. The company also provides an example of an agent connecting through Chrome DevTools MCP to a Browser Run endpoint and listing available WebMCP tools. These examples show where an agent can discover and call a page’s declared functions; they are not a measured comparison of completed customer tasks.[1]
The change only provides a direct tool path when a website exposes suitable functions. Pages without them still need other forms of navigation, and a tool’s appearance alone does not prove that an entire booking or form submission can finish. Cloudflare also describes wider CSS, module and rendering support in the browser update, but these are compatibility changes alongside the WebMCP addition. The reportable new capability is Kitesurf’s support for calling page tools.[1]
