browser-switch

name	browser-switch
description	Choose between agent-browser and Playwright CLI for browser automation tasks.

Browser Switch

Choose between agent-browser (Rust) and Playwright CLI (Node.js) for browser automation.

Quick Choice

Need max speed & token efficiency? → agent-browser Need cross-browser or debugging? → Playwright CLI Complex component frameworks (e.g. Gradio, MUI, Ant Design)? → Playwright CLI

When to Use agent-browser

• Token efficiency matters (5.7x better than Playwright)
• Speed is critical (~100ms per action)
• Chrome-only is acceptable
• AI agent workflows where tokens are limited
• Rust performance preferred
• Simple web pages with standard HTML elements

When to Use Playwright CLI

• Need Firefox, Safari, or Edge testing
• Complex auto-waiting/retry logic required
• Already invested in Playwright ecosystem
• Better debugging and tracing needed
• Formal testing scenarios
• Complex component frameworks (e.g. Gradio, MUI, Ant Design)
• Tab/dialog state must persist across navigation

Key Differences

Feature	agent-browser	Playwright CLI
Language	Rust	Node.js
Browser	Chrome only	Chrome, Firefox, Safari
Token Usage	~5.5K chars	~31K chars
Speed	~100ms/action	~200-300ms
Maturity	New (Vercel)	Mature (Microsoft)
Custom components	May miss some	Better snapshot coverage

Examples

AI Agent Task (agent-browser)

agent-browser open example.com
agent-browser snapshot -i
agent-browser click @e2
agent-browser fill @e3 "[e-mail]"

Cross-Browser Test (Playwright CLI)

playwright-cli open https://example.com
playwright-cli snapshot
playwright-cli click e2
playwright-cli fill e3 "[e-mail]"

Decision

Use agent-browser unless you specifically need:

• Cross-browser testing
• Advanced debugging features
• Playwright ecosystem compatibility
• Complex component frameworks (e.g. Gradio, MUI, Ant Design)

Troubleshooting

agent-browser can't find an element

If agent-browser snapshot -i doesn't show the element:

1. Try CSS selector: agent-browser click "#my-button"
2. Try text search: agent-browser find text "Submit" click
3. Try JavaScript: agent-browser eval "document.querySelector('.my-class').click()"
4. Fall back to Playwright CLI for better snapshot coverage

Tab/dialog state lost on navigation

agent-browser may lose tab state when navigating away and back. Use CSS selectors for tab switching:

agent-browser click "[role=tab]:nth-child(2)"

Or fall back to Playwright CLI for reliable tab persistence.

Complex component frameworks

Apps built with Gradio, MUI, Ant Design, or other component libraries often use custom elements and deep nesting that agent-browser's accessibility tree doesn't fully capture. Common symptoms:

• Tabs, modals, or drawers appear - text: "..." with no interactable refs
• Custom dropdowns, selects, or sliders are invisible in the snapshot
• API documentation popups or settings panels don't appear at all

Try Playwright CLI when agent-browser struggles with these patterns.