Puppeteer Real Browser MCP Server

# Test Instructions for Resilient get_content ## Test Scenario: cursor.com Login Flow ### Step 1: Initialize Browser ``` browser_init ``` ### Step 2: Navigate to cursor.com ``` navigate with url: "https://www.cursor.com/" ``` ### Step 3: Use get_content (will auto-retry) ``` get_content ``` Expected behavior: - First attempts with contentMode="main" - If that exceeds token limits, automatically retries with contentMode="summary" - Should succeed with summary mode and show: "📊 Content Mode: Automatically used 'summary' mode" ### Step 4: Find Login selector ``` find_selector with text: "Login" ``` Expected behavior: - Should now work because content was successfully analyzed in summary mode - Returns a valid selector for the login element ### Step 5: Click Login ``` click with the selector returned from step 4 ``` ## Key Improvements 1. **Automatic Retry**: get_content now automatically tries progressively smaller content modes 2. **Strict Validation**: find_selector still requires successful content analysis (no blind guessing) 3. **Better UX**: Users don't need to manually retry with different modes 4. **Clear Feedback**: System tells you which mode was used automatically ## Testing Other Large Sites Try these sites that are known to be large: - github.com - stackoverflow.com - reddit.com - youtube.com The system should automatically adapt to their size by using appropriate content modes.