Use Cases

Evidence capture

Use get_page_snapshot to record URL, title, and visible text, then take_screenshot for a PNG artifact. Finish with list_artifacts or get_file to retrieve the evidence bundle.

Best for QA checks, before/after captures, audit notes, and bug reports.

Paginated scraping

Use navigate, get_page_snapshot, click, scroll, and wait_for to move through search results, tables, product lists, or article lists when there is no API available.

Keep a page limit and stop on login walls, CAPTCHA, consent walls, duplicate content, or missing next controls.

Standard login forms

Use navigate, get_page_snapshot, type_text, click, press_key, and wait_for for ordinary email/password forms. Do not echo passwords, TOTP codes, cookies, or auth headers in responses or artifacts.

This is not for SSO, OAuth, passkeys, CAPTCHA solving, or bypassing security controls.

Multi-step browser tasks

Use browse_and_act_structured when the task is easier to describe as an outcome than as a list of selectors. The agent starts from the provided URL and uses the configured OpenAI-compatible model to complete the instruction.

Use direct tools instead when the flow needs deterministic selector-by-selector control.

Screenshot reports

Use take_screenshot with full_page: true for long reports or selector for a focused element. Retrieve files from the host at ${DATA_PATH}/browser-mcp or through get_file.