Kitesurf – Cloudflare’s Lightweight Cloud Browser for AI Agents Operational

Kitesurf is a lightweight cloud browser launched by Cloudflare specifically for AI Agents, running in the Workers V8 isolate environment. Built in Rust, it removes redundant human-oriented features such as tabs and extensions while retaining DOM parsing, JavaScript execution, and data extraction capabilities. Compared with Chromium, it reduces CPU usage by more than 3 times and memory usage by 4.7 to 7 times. It supports CDP and MCP, allowing Puppeteer and Playwright to connect directly.

Lightweight web rendering: Built in Rust, Kitesurf supports HTML parsing, CSS, JavaScript execution, and DOM manipulation, and can correctly render real websites such as TodoMVC, Wikipedia, and Hacker News.
CDP compatibility: It fully supports the Chrome DevTools Protocol, allowing existing tools such as Puppeteer, Playwright, and chrome-remote-interface to connect without rewrites.
MCP support: AI Agents can control Kitesurf instances directly through the Model Context Protocol for tool calling and automation.
Quick Actions: Use Browser Run’s Quick Actions API to capture screenshots, extract HTML, or generate PDFs with a single request by adding the browser=kitesurf parameter.
Online Playground: A public playground lets you view page rendering, DOM structure, network activity, and memory usage in real time.

Quick Actions: Add the browser=kitesurf parameter to a Browser Run API request to directly perform stateless tasks such as screenshot capture, HTML extraction, or PDF generation.
Connect existing tools through CDP: Append ?browser=kitesurf to the URL of Browser Run’s WebSocket CDP endpoint. Existing clients such as Puppeteer and Playwright can then connect seamlessly and run automation scripts.
Let AI Agents control it through MCP: Enter Kitesurf’s WebSocket endpoint and API Token in your MCP client configuration. AI Agents can then control the browser directly through the standard MCP protocol for browsing and data extraction.
Try it online in the Playground: Visit https://kitesurf.cloudflare.app/ and enter any URL to view page rendering, DOM structure, network activity, and memory usage in real time.

Highly resource-efficient: Compared with Chromium, screenshot tasks use 3.1 times less CPU and 4.7 times less memory; HTML extraction tasks use 3.8 times less CPU and 7.0 times less memory.
Request-level statelessness: Each task starts, runs, and is destroyed independently, making Kitesurf naturally suited to high-concurrency and bursty Agent workloads.
Edge deployment: Powered by Cloudflare Workers and its global edge network, browser instances can run close to users to reduce latency.
Strong standards compatibility: Kitesurf has passed more than 235,000 WPT subtests, with scores of 97% for DOM, 96% for HTML, 99% for Selection, and 99% for Encoding.
Minimal migration cost: With compatibility across both CDP and MCP, developers do not need to learn a new API, and existing automation scripts can switch runtimes seamlessly.

AI Agent web search and information extraction: Agents can visit webpages in batches and extract structured data such as titles, main text, and links for RAG or knowledge base construction.
Batch web automation: Headless tasks that need to process many pages simultaneously, such as e-commerce price monitoring, inventory checks, and competitor data collection.
Headless screenshots and PDF generation: Quickly generate screenshots or PDFs for compatible sites for reporting, archiving, or content review.
MCP-driven AI workflows: AI Coding Agents or research Agents can control the browser directly through MCP to log in, fill out forms, and collect data.
Short-lived edge rendering: Quickly render pages and return results from global edge nodes for Serverless scenarios that do not require long-lived session state.