Browser Use

• Web scraping and data extraction: You need to pull structured information from pages that require interaction (login, pagination, dynamic rendering) before the data is visible.
• Automated form filling and submission: You need to fill out and submit web forms, including multi-step wizards that depend on prior page state.
• End-to-end web testing: You want to exercise a web app's UI in a real browser environment, capturing screenshots and verifying element states.
• Authenticated browsing workflows: You need to operate inside sites you're already logged into by syncing local Chrome profiles or reusing cloud sessions with saved cookies.
• Parallel cloud AI agents: You need to dispatch multiple autonomous browser agents concurrently (e.g. competitive research, bulk data collection) and poll their results.

• Simple HTTP API calls: If the site exposes a REST or GraphQL API, calling it directly is faster and more reliable than browser automation.
• Static HTML scraping: If the page doesn't require JavaScript rendering or interaction, a lightweight HTTP fetch is more efficient.
• Non-browser automation: Tasks that don't involve a web browser (file system operations, database queries, etc.) are out of scope for this skill.

The browser stays open between commands, so you can build multi-step workflows with open, state, click, input, and screenshot without restarting. Named sessions (--session NAME) let you run multiple independent browsers in parallel.

Choose between chromium (fast, headless, isolated), real (your installed Chrome binary with optional profile and cookies), or remote (cloud-hosted browser with proxy support). Each mode is suited to different tradeoffs between isolation, authentication, and cost.

List, inspect, sync, import, and export browser profiles and cookies — locally or in the cloud. Domain-scoped cookie syncing lets you carry authentication from your local Chrome profile into a cloud browser session without exposing unrelated session tokens.

Launch autonomous browser agents in the cloud with browser-use -b remote run "task". Supports async execution, parallel sessions, session reuse across sequential tasks, structured JSON output, LLM model selection (e.g. gpt-4o, claude-sonnet), and a judge mode for evaluation.

Run arbitrary Python code inside the browser session with browser-use python. Variables and the browser object persist across commands, giving you programmatic access to navigation, interactions, and page data without leaving the CLI.

Expose a local dev server to cloud browsers with browser-use tunnel <port>, which creates a Cloudflare tunnel and returns a public URL. Tunnels are independent of browser sessions and persist across close commands.

Sync cookies from your local Chrome profile, open a cloud session authenticated to your SaaS tool, navigate to the export page, click the download button, and wait for completion — all without manual login.

Launch three simultaneous cloud agent tasks each targeting a different competitor's pricing page. Poll their status with task status, collect the structured output when each finishes, and stop all sessions at once with session stop --all.

Open your local dev server via a tunnel, walk through a multi-step form using element indices from state, capture screenshots at each stage with browser-use screenshot path.png, and verify final page text with browser-use wait text "Success".

Navigate to a JavaScript-rendered listing page, scroll down to load more results, use browser-use get html --selector ".results" to extract the relevant fragment, and run browser-use eval to pull structured data out of the DOM.