browser-use

# Browser Use ## Purpose Drive an interactive browser for exploration and scripted UI flows: open pages, act on candidates, and return observations from tool output—without inventing selectors or page state. ## When to Use - Prototype or experimental browser automation agents. - Navigate, click, type, or wait on dynamic pages through browser-use / agent-browser style tools. - Capture page state after actions for follow-up reasoning. ## When NOT to Use - Product demo recording on the dedicated demo computer → **product-demo** / `demo_*` tools when that family is equipped. - Cloud Browserbase MCP sessions → **browserbase**. - Static HTTP fetch without rendering. ## Expected Outcome - Session started before interaction. - Actions use tool-returned candidates or explicit approved selectors—never invented CSS/XPath. - Results summarized from tool responses only. ## Workflow 1. Confirm browser automation tools are available for the session. 2. Start or reuse a browser session; navigate to the target URL. 3. Observe returned candidates / snapshot; pick ids from the latest observation. 4. Act (click/type/wait) using only ids or hrefs from that observation. 5. Summarize outcomes; escalate when auth, MFA, or CAPTCHA blocks progress. ## Main tools Prefer platform `agent-browser` / remote browser / browser-use MCP tools exposed to the agent. Do not invent tool names. ## Safety and Boundaries - Do not bypass paywalls, CAPTCHAs, or MFA. - Do not store or echo secrets from pages. - Prefer read-only exploration unless the user approved writes.