Agent Browser

Most browser automation tools are built for humans running test suites. Agent Browser is built for AI agents running autonomously.

Powered by Vercel's agent-browser CLI, it uses accessibility snapshots and ref-based element selection (@e1, @e2) — no CSS selectors to maintain, no XPath fragility, no Playwright dependency.

How it works:

• Take a snapshot of the page → get a list of interactive elements with refs
• Act on elements by ref: click @e3, fill @e7 with text
• Navigate, screenshot, and scrape without writing selector logic

What you get:

• Setup check: verify agent-browser CLI is installed
• Snapshot-first workflow: always snapshot before acting
• Core operations: navigate, click, fill, select, screenshot, scrape
• Form completion patterns: multi-step forms, dropdowns, checkboxes
• Error handling: element not found, navigation timeout, stale refs
• Pagination and infinite scroll patterns

vs. CDP Browser Automation: Agent Browser uses accessibility tree refs (simpler, more stable). CDP gives lower-level control. Use Agent Browser for most tasks; use CDP when you need JavaScript execution or precise pixel coordinates.

Requires: npm install -g @vercel/agent-browser and a running browser.