Browser Automation for AI Coding Agents

agent-browser talks to Chrome directly over the DevTools Protocol without a heavy browser framework. Its core loop relies on accessibility tree snapshots that assign short refs (@e1, @e2) to interactive elements, keeping LLM token overhead minimal.

It provides commands for navigation, form filling, screenshots, and state persistence across multi-step browser tasks.

• Accessibility tree snapshots keep prompt context small
• Direct CDP CLI with no Playwright runtime overhead
• Parallel sessions for multi-agent workflows
• Official Vercel Labs skill with 728K+ registry installs

Anthropic's official skill for local web app QA. It guides agents to write native Python Playwright scripts using a reconnaissance-then-action workflow to test dynamic frontend features.

Includes a bundled with_server.py helper that handles starting and stopping local development servers around test execution.

• Official Anthropic skill for UI verification
• Automatic server lifecycle management during test runs
• Reconnaissance-then-action pattern for dynamic DOMs
• Captures screenshots and console logs for bug reporting

Maintained by Microsoft's Playwright team, this skill wraps the Playwright engine into single-purpose CLI commands for clicking, typing, and navigating.

Inherits Playwright's auto-waiting and cross-browser engine features while using element refs for compact LLM prompts.

• Maintained by the Playwright team at Microsoft
• Built-in auto-waiting and cross-browser support
• Integrates cleanly with existing Playwright test suites
• Element ref snapshot interface

A comprehensive reference skill by Currents covering locators, fixtures, Page Object Model, flaky test debugging, and CI optimization.

Provides contextual guidance for complex scenarios like WebSockets, accessibility testing, and visual regression.

• Activity-based index across 40+ Playwright reference guides
• Deep guidance on flaky tests and async wait patterns
• Covers visual regression, accessibility, and API mocking
• Written by Playwright CI observability engineers at Currents

browser-use allows agents to write Python snippets that execute against a running Chrome browser, preserving existing logins and session cookies.

Features a built-in doctor command for environment connection diagnostics and site-specific domain pattern libraries.

• Attaches to your existing Chrome profile with active sessions
• Python heredoc interface with pre-imported helpers
• Built-in connection diagnostic doctor command
• Site-specific automation pattern modules

browser-skill automates the user's logged-in Chromium browser through an Agent Window. Its primary workflow covers reading pages, filling forms, operating tabs, testing a UI, and debugging a website with the user's existing logins.

The skill requires the bsk CLI and browser extension. It makes session ownership explicit, requires fresh page observations before interaction, and returns borrowed tabs when the session stops.

• Uses a logged-in Chromium browser in an Agent Window
• Reads pages, fills forms, operates tabs, and tests a UI
• Explicit borrowing and return of user tabs
• Fresh page refs before browser interaction

browser-act prioritizes privacy and session safety through its Confirmation Gate protocol, requiring user approval before executing sensitive browser operations.

Supports isolated multi-browser profiles for parallel account management alongside network HAR/XHR capture capabilities.

• Confirmation Gate protocol requires explicit approvals
• Strict local-only storage for credentials and cookies
• Isolated multi-browser profiles for parallel account work
• Network waterfall capture (HAR, XHR, fetch)

Browserbase's skill drives the browse CLI, enabling switching between local Chrome instances and remote Browserbase cloud sessions.

Remote cloud sessions provide stealth features like automatic CAPTCHA solving, residential proxies, and Browserbase Identity verification.

• Switch between local Chrome and cloud sessions per command
• Automatic CAPTCHA solving and residential proxy support
• Attach to active debuggable Chrome instances
• Designed for bot-protected and complex targets

Official MCP server from the Chrome DevTools team giving agents deep access to performance traces, network request inspection, and heap snapshots.

Encourages routing large trace files and diagnostic outputs directly to disk to prevent context window pollution.

• Built and maintained by the Chrome DevTools team
• Performance profiling, network waterfalls, and memory dumps
• Disk output routing keeps large traces out of context
• Includes sibling skills for LCP and accessibility debugging