cursor-ide-browser-skill_skill

Overview

This skill provides browser automation inside Cursor IDE using the MCP server cursor-ide-browser and accessibility snapshots for precise element control. It streamlines navigation, interaction, screenshots, OCR processing, and snapshot querying so you can automate web tasks reliably. The workflow emphasizes taking an accessibility snapshot before any interaction to obtain stable element refs.

How this skill works

Navigate to a page, call browser_snapshot() to capture an accessibility tree (YAML) that includes element refs, then use those refs for clicks, typing, selection, and other actions. Optionally convert snapshots to Markdown and grep the file for fast element discovery or run snapshot-query tools (find_by_name, find_by_role, regex, BM25) to programmatically extract refs. Use browser_take_screenshot() + fast-paddleocr-mcp for OCR-based extraction and then query OCR snapshot logs the same way.

When to use it

• Automating navigation, clicks, form fills, and link interactions within Cursor IDE
• Scraping or extracting structured information from pages where element refs improve reliability
• Testing web application flows (login, search, pagination) with reproducible refs
• Capturing visual evidence with screenshots and debugging using console and network logs
• Processing page images with OCR when text is non-selectable or rendered as images

Best practices

• Always call browser_snapshot() after navigation or page changes — refs are page-state specific
• Convert snapshots to Markdown and grep for fast, human-friendly element discovery
• Prefer snapshot-query BM25 for semantic searches when exact labels aren’t available
• Wait for dynamic content with browser_wait_for() before interacting or taking screenshots
• Re-generate snapshots after significant DOM updates or navigation to avoid stale refs

Example use cases

• Automated login: navigate → snapshot → type username/password by ref → click login → wait for welcome
• Search-and-extract: navigate to results → snapshot → convert to Markdown → grep for keywords → click result
• Visual extraction: take screenshot → OCR via fast-paddleocr-mcp → query OCR snapshot for measurements or labels
• Debugging: snapshot → browser_console_messages() and browser_network_requests() to surface errors and failed calls
• Multi-source verification: navigate multiple pages, snapshot each, extract and compare facts for consistency