The Pulse

Cloudflare open-sources experimental Kitesurf browser for AI agents

Developers can now inspect, modify and run Cloudflare's Kitesurf browser engine locally. Deploying an independent instance on Cloudflare requires a Workers Paid plan.

AI.info Team · · fact-checked · against 3 sources ↓ · Published 13 Nov 2026

Developers can now inspect, modify and run Cloudflare's Kitesurf browser engine for AI agents after the company released its code under the Apache 2.0 license. The experimental engine can also be deployed to a developer's own Cloudflare account, according to a company post published November 13.

Running a self-deployed instance on Cloudflare requires a Workers Paid plan. Kitesurf uses Dynamic Workers to create isolated Workers for pages; developers can also run the engine locally.

A browser built for agents

Kitesurf runs on Cloudflare Workers rather than in a browser container. The browser parses HTML, applies CSS, executes page JavaScript and renders screenshots. It supports the Chrome DevTools Protocol, allowing connections from clients such as Puppeteer.

Cloudflare's stated goal is a browser that agents can use to load and interact with pages while consuming less CPU and memory than a full browser such as Chromium. The engine remains under heavy development, with compatibility, reliability and efficiency among the areas the team is working on.

How the engine is assembled

Kitesurf has three main components: the Engine manages browser sessions and client connections, PageScript runs the page and its JavaScript, and PageRenderer draws the result. The repository separates those pieces from other packages, including the Rust core, DOM shims, renderer and command-line interface.

The project combines Cloudflare platform APIs with open-source Rust packages compiled to WebAssembly. Blitz handles HTML parsing, the document object model, layout and painting, while Stylo provides CSS parsing, cascading and computed styles.

Boa executes JavaScript inside WebAssembly when it cannot run in Dynamic Workers. Linebender libraries handle text layout and 2D graphics, and Krilla generates PDF output. The individual projects keep their respective licenses, with third-party notices documented in the repository.

Running an instance

The first local build compiles the Rust core and its dependencies; later builds reuse the build cache. A local development server then allows a page to be opened in Chrome DevTools or used programmatically, including through a Puppeteer client or an AI harness connected to the local protocol port.

Self-deployment provides an alternative to Cloudflare's managed Browser Run APIs. The repository warns that the engine has no authentication of its own: someone with its URL can connect clients, browse through the engine, send outbound requests from the account and use resources the account pays for. It advises placing an authentication layer in front of a deployed instance.

Earlier updates

Kitesurf was introduced in August, followed by an update in September that added WebMCP support, APIs and compatibility improvements. WebMCP lets website developers expose functions directly to agents, which can call them rather than simulate clicks.

The September update also extended Browser Run API coverage to Kitesurf, including access through Playwright and MCP. At that time, Cloudflare said the engine passed more than 730,000 Web Platform Tests subtests, a measure of support for web standards.

For contributions, Cloudflare asks developers to open an issue with a minimal reproduction or proposed approach and obtain a maintainer's approval before opening a pull request. Bug reproductions, failing automation workflows and documentation corrections can also be contributed.