Cloudflare Kitesurf: Cloudflare has published Kitesurf, its experimental browser for AI agents, under the Apache 2.0 license. The November 12 changelog and November 13 blog announcement confirm that developers can inspect the code, run it locally and deploy their own instance on Workers. Cloudflare Kitesurf is designed for browser automation without a full browser container. The release is useful for developers researching agent infrastructure, but it does not make the engine a drop-in replacement for Chromium or make production deployments free.

A browser engine for agent workflows
An agent connects through CDP, the engine processes a page on Workers, and the site can expose supported structured tools. A deployed instance needs authentication. The illustration is a high-level model, not a complete architecture or measured test.
What Happened
Cloudflare introduced Kitesurf in August and subsequently added WebMCP support in September. With the open-source release, developers can inspect and modify the implementation instead of relying only on a hosted service. Cloudflare Kitesurf parses web pages, applies CSS, runs page JavaScript and produces screenshots on Cloudflare Workers. It exposes a Chrome DevTools Protocol endpoint so compatible clients such as Puppeteer can connect.
The engine is not a full Chrome implementation, and the project documentation explicitly warns that browser standards support, rendering fidelity and performance are still under development. The release makes the code available, not every browser behavior reliable.
Who Is Affected & When
The release targets developers building AI browser automation, testing workflows and research agents. Local experimentation is possible with Wrangler and the repository build instructions. Running a self-hosted Cloudflare Kitesurf instance in the Cloudflare network requires the Workers Paid plan because the engine uses Dynamic Workers to create isolated page workers. The repository also states that deployed endpoints need protection through Cloudflare Access or another authentication layer.
The license is permissive, but the operational cost of hosting, resource usage and maintenance still belongs to the operator. Teams should distinguish this self-deployed engine from the managed Cloudflare Browser Run service and review the relevant product documentation.
What Changes in Practice
1. Cloudflare Kitesurf and browser automation
A conventional headless browser automation setup normally drives a browser process through a client library. Kitesurf instead implements browser functions inside Workers and exposes CDP for external clients. That makes it an interesting option for tasks such as opening a public page, reading rendered text or collecting screenshots. A team should not assume that existing Puppeteer tests will run without changes: CDP compatibility is incomplete.
Start with a small set of read-only tasks and compare outputs with a known-good browser. If you already use a browser automation framework, isolate the adapter for Kitesurf so a failed task can fall back to your existing browser.
2. WebMCP and an AI agent browser
WebMCP lets a compatible browser expose structured tools registered by a website, giving an agent a way to call defined functions rather than infer every action from the page layout. Cloudflare added WebMCP support to Kitesurf before the open-source release. WebMCP and a remote MCP server are distinct interfaces: the former exposes site tools to a browser context, while the latter is a network service an MCP client can call. Neither makes a tool inherently safe.
Verify names, arguments, results and permissions, and require human approval for meaningful external actions.
3. Local setup and the Chrome DevTools Protocol
The repository lists Bazelisk, Node.js 22 or newer, and a pinned pnpm version among the build requirements. On macOS it also specifies GNU tar. Follow the current repository instructions, install dependencies with the frozen lockfile, then run pnpm run dev. This builds the engine and starts Wrangler locally; the documented health endpoint is 127.0.0.1:8787/health. CDP clients can connect to the documented local WebSocket endpoint.
These are vendor-provided steps, not a claim that DigiOffly has completed a Kitesurf build or verified every dependency.
4. Workers Paid, Cloudflare Access and deployment costs
The code is Apache 2.0 licensed, but running Kitesurf on Cloudflare requires Cloudflare Workers Paid and Cloudflare Dynamic Workers. Treat the workload as a metered application, not a free unlimited remote browser automation service. Review the current Workers pricing and usage limits before deploying. Cloudflare Kitesurf has no built-in authentication for its own deployed endpoint.
The repository warns that an exposed URL can let outsiders browse using your engine, make outbound requests from your account and consume paid resources. Protect the deployment with Cloudflare Access before sharing it. Use explicit budgets, access policies and an allowlist of permitted tasks.
5. A controlled pilot for browser automation testing
Pick three safe, representative tasks: open an editorial article, extract a public product title and capture a screenshot of a responsive page. Run them in the existing browser and Cloudflare Kitesurf with identical inputs. Compare the visible text, important UI elements, output correctness and failure modes. Include a page with a JavaScript widget and a page with a structured WebMCP tool if those are relevant to the workflow. Record wall-clock time and observed costs only after real runs.
A small reproducible pilot is more useful than a broad speed claim based on an untested engine.
| Decision | Kitesurf | Pilot requirement |
|---|---|---|
| Engine | Workers-native experimental browser | Check HTML, CSS and JS output |
| Client interface | CDP; compatible clients such as Puppeteer | Test required commands |
| Website tools | WebMCP support announced earlier | Check tool discovery and execution |
| Deployment | Local Wrangler or Workers Paid | Review cost and quotas |
| Security | No built-in endpoint authentication | Enforce access controls and budgets |
Why It Matters
AI agent browser tools are becoming a separate infrastructure choice. Developers can now inspect how Cloudflare Kitesurf handles web content and decide whether the approach fits a particular job. For simple discovery and public research, an agent may not need all the behavior of a desktop browser. For complex applications, authenticated workflows or visual quality assurance, browser compatibility and reliable state handling can matter more than deployment convenience.
Open-source access helps teams evaluate these trade-offs without accepting a marketing comparison as a benchmark. For agentic browser automation, that means choosing a runtime based on the operations your agent must perform, rather than assuming one engine is best for every site.
Our Take
DigiOffly has already made its public software research resources available through its own read-only MCP endpoint and WebMCP registrations. Our first sensible Kitesurf experiment would be to open a public DigiOffly page and call a software discovery tool. We would compare the result with the existing Chrome-based workflow and test that private functions remain inaccessible. This is a proposed test, not a claim that we have installed or benchmarked Kitesurf.
Keep our production site, WordPress administration and trading terminal outside the pilot. If the engine handles the public tasks correctly, it could become one option in a multi-agent research system. If compatibility is incomplete, use the existing browser for that task. Cloudflare Kitesurf is promising, but reliable results, access controls and a measured budget should determine whether it belongs in production.
Frequently Asked Questions
Is Cloudflare Kitesurf free and open source?
Its code is available under Apache 2.0. Cloudflare-hosted self-deployment requires Workers Paid because the engine uses Dynamic Workers. Hosting usage is not made free by the license.
Can I use Puppeteer with Cloudflare Kitesurf?
Yes, Kitesurf supports CDP and documents connecting compatible clients such as Puppeteer. It is not a complete Chrome implementation, so test the features you need.
Does Kitesurf support WebMCP?
Cloudflare announced WebMCP support in its September update. Use a compatible site and verify that its specific tool registrations and calls behave correctly.
Is Cloudflare Kitesurf ready for production?
The project remains experimental. Protect any deployed endpoint with authentication, test compatibility and reliability, and review costs before relying on it for production tasks.
Official Sources & Dates
Announcement: 2026-11-13; effective or availability date stated by the vendor: 2026-11-12. Check official updates before purchasing or changing production settings.
- Cloudflare: Kitesurf is now open source
- Cloudflare changelog: open-source release and requirements
- Cloudflare Kitesurf repository and setup instructions
- Cloudflare Kitesurf project site
- Cloudflare Workers pricing
Editorial note: DigiOffly verified dated primary sources. This is not a claim of hands-on testing unless independently documented. Prices and access conditions may change.