Browser Automation

Available Tools

• browser_navigate(url): Navigate to a URL and capture screenshot
• browser_act(instruction): Execute browser actions using natural language (click, type, scroll, select). Does NOT support drag.
• browser_extract(extraction_instruction): Extract structured data (auto-scrolls entire page, collects all data in single call)
• browser_get_page_info(): Get page structure and all open tabs (fast, no AI)
• browser_manage_tabs(action, tab_index): Switch, close, or create browser tabs
• browser_save_screenshot(filename): Save current page screenshot to workspace

Tool Selection

• browser_navigate + browser_act: UI interactions (click, type, scroll, form fill)
• browser_extract: Structured data from visible content (auto-scrolls)
• browser_get_page_info: Fast page structure check (<300ms)
• browser_save_screenshot: Save milestone screenshots (search results, confirmations, key data)

browser_act Best Practice

• Combine up to 3 predictable steps: "1. Type 'laptop' in search 2. Click search button 3. Click first result"
• On failure: check the screenshot to see current state, then retry from that point
• For visual creation (diagrams, drawings), prefer code/text input methods over mouse interactions