agent-browser

What is agent-browser

agent-browser is a browser automation CLI tool for AI Agents launched by Vercel Labs. It is written in native Rust and directly controls a real Chrome browser based on the Chrome DevTools Protocol (CDP). Through a compact accessibility tree Snapshot and Ref reference mechanism, the tool provides LLMs with context-efficient web interaction capabilities, supporting 50+ browser operation commands, multi-session isolation, cloud browser access, and a real-time observability Dashboard.

Main features of agent-browser

• Snapshot accessibility tree capture: Outputs a compact web page accessibility tree structure, with deterministic Ref references such as @e1、@e2 , for AI to precisely locate elements.
• Ref-driven interaction: Executes operations such as clicking, filling, scrolling, and screenshotting based on the Ref references returned by the Snapshot, without the need to repeatedly query the DOM.
• 50+ browser operation commands: Covers complete browser control capabilities such as navigation, forms, mouse, keyboard, clipboard, network interception, tab management, and PDF export.
• Semantic element location: Supports finding and operating elements through semantic methods such as ARIA roles, text content, labels, placeholders, and alt text.
• Multi-session isolation: Supports running multiple browser sessions in parallel, each with independent cookies, storage, authentication, and navigation history.
• Network monitoring and interception: Can intercept, block, and mock network requests, supports HAR recording, and filtering requests by resource type, status code, and method.
• Screenshots and visual Diff: Supports full-page/viewport screenshots, screenshots with numbered annotations, pixel-level visual comparison, and Snapshot text Diff.
• React / Web Vitals introspection: Built-in React DevTools hooks can output the component tree, Fiber render records, Suspense boundaries, and performance metrics such as LCP/CLS/TTFB.
• Real-time streaming and Dashboard: Pushes the browser viewport in real time via WebSocket, and the local Dashboard provides an activity stream, console output, and an AI Chat panel.
• AI natural language control: Integrates with Vercel AI Gateway, supports agent-browser chat to drive the browser to complete complex tasks with natural language instructions.

How to use agent-browser

• Install the CLI: Execute npm install -g agent-browserto globally install the native Rust binary.
• Initialize the browser: Run agent-browser install to automatically download Chrome for Testing, or reuse an existing Chrome/Brave installation.
• Open the target page: Execute agent-browser open example.com to launch the browser and navigate to the target URL.
• Get the page Snapshot: Run agent-browser snapshot -i to output the interactive element accessibility tree and Ref references.
• Perform interactive operations: Use Ref references to execute operations such as agent-browser click @e2、agent-browser fill @e3 "test@example.com" .
• Screenshot or export: Execute agent-browser screenshot page.png or agent-browser pdf report.pdf to save the result.
• Close the session: Run agent-browser close to end the current browser session.

Core advantages of agent-browser

• AI Context Efficiency First: Compact text output drastically reduces LLM context usage compared to JSON or full DOM, designed specifically for Agent workflows.
• Real Browser Anti-Detection: Operates a real Chrome via CDP instead of a headless browser, naturally carrying user profiles and reducing the probability of being blocked by risk controls.
• Zero Node.js Dependencies: Pure Rust native binary, no Playwright or Node runtime required, lightweight installer and extremely fast startup.
• Cross-Platform Native Support: Provides native compiled versions for all platforms: macOS (ARM64/x64), Linux (ARM64/x64), Windows (x64).
• Enterprise-Grade Security Mechanisms: Offers optional security layers including encrypted authentication vault, content boundary markers, domain whitelist, action policy files, sensitive operation confirmation, and output length limits.
• Seamless Cloud Browser Integration: Supports one-click integration with cloud browser infrastructure such as Browserless, Browserbase, Browser Use, Kernel, AWS AgentCore, and supports Serverless deployment.

agent-browser project address

• Project Website：https://agent-browser.dev/
• GitHub Repository：https://github.com/vercel-labs/agent-browser

Comparison of agent-browser with similar competitors

agent-browser use cases

• AI Agent web automation: Enabling LLMs to autonomously complete end-to-end web tasks such as information retrieval, form filling, data scraping, and competitor monitoring.
• End-to-end testing and regression: Using Snapshot Diff and screenshot pixel comparison for UI regression testing, combined with semantic locators to reduce test fragility.
• Headless data collection: Executing structured data scraping and report generation in CI/CD or Serverless environments (Vercel, AWS Lambda).
• Mobile web testing: Controlling Mobile Safari via iOS simulators or real devices to verify mobile page compatibility and interaction flows.
• Frontend performance monitoring: Automatically collecting Web Vitals metrics and React component rendering data, integrated into monitoring pipelines.