AgentCore Browser Web Scraping

Amazon Bedrock AgentCore Browser gives you a managed cloud Chromium — no crawler fleet, no resident containers, per-session billing, natural isolation. This skill captures a production scraping architecture built on it: an LLM agent decides what to do on the page, but only through fixed code primitives it can parameterize, never arbitrary scripts. Login state lives in Browser Profiles that users populate once through a live-view session, so credentials never transit the conversation.

Collector (Python, worker thread)
  ├─ browser_session(region, profile_configuration=..., proxy_configuration=...)
  │      └─ StartBrowserSession → generate_ws_headers() (SigV4)
  │             └─ playwright chromium.connect_over_cdp(ws_url, headers)
  └─ Strands Agent (LLM) with 6 fixed tools:
        navigate / scroll_to_bottom / click_load_more /
        get_page_text / screenshot / extract_by_selector