docs: explain live V4 network capture over CDP

[MagMueller]

What changes

Adds a V4 guide for attaching a CDP observer to an existing live browser. It explains how the browser ID differs from run/session IDs, how to retrieve cdpUrl, and how to save bounded, origin-only request metadata with Playwright. Links the page from Observability and navigation, and regenerates the four Cloud/root LLM-text mirrors.

The guide explicitly does not promise complete historical capture or HAR export. It covers late attachment, existing-page-only coverage, redirects, browser replacement, sensitive data, and the difference between detaching a logger and stopping a browser. No SDK or runtime behavior changes.

Verification

• Live Cloud test: created one disposable V4 agent run through the signed-in Cloud UI. A separate CDP client attached to its browser and recorded requestWillBeSent, responseReceived (HTTP 200), and loadingFinished for a controlled same-origin GET. No customer session, login, or form submission was involved.
• Important limit: the stored CLI API credential returned HTTP 403 for V4 API calls. The browser-discovery REST path is checked against the current API schema/source, but was not verified end to end with that credential. The live test used the disposable run's existing live-view browser connection. This does not prove early/startup request coverage, every target, or a historical export.
• Executed the exact Python example against a synthetic CDP transport: four event types, page/request correlation, omission of URL paths/queries/userinfo/headers/bodies, owner-only output permissions, refusal to overwrite, and client detachment passed. This is separate from the live test above.
• mint broken-links: passed.
• Regeneration is idempotent; Cloud/root mirrors match.
• git diff --check: passed.
• Inspected the Mintlify page at desktop width and 390px; no document-level horizontal overflow.

The real test run completed. Its managed browser remains subject to the normal agent lifecycle; an explicit stop through the API was not verified with the available credential. The example does not create or stop browsers.

---

Summary by cubic

Adds a V4 guide for attaching a CDP observer to a live browser and saving bounded, origin-only request metadata. The page explains how browser IDs differ from run/session IDs, how to retrieve cdpUrl, and covers late attachment, redirects, browser replacement, and sensitive data; run events are explicitly not a network log or a historical HAR export. The example distinguishes missing BROWSER_CDP_URL configuration from connection failures, and connection failures raise a generic message so the credential-bearing CDP URL never appears in tracebacks.

• Links the guide from the Observability page and the navigation sidebar.
• Regenerates the Cloud and root LLM-text mirrors.
• Requests are keyed by (page, requestId); recorded origins keep IPv6 brackets and non-default ports while paths, queries, headers, cookies, and bodies are dropped, and malformed URLs record a null origin.

^{Written for commit 9427bbe. Summary will update on new commits.}

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated
browser-use	🟢 Ready	View Preview	Sep 5, 2026, 6:19 AM

💡 Tip: Enable Automations to automatically generate PRs for you.

[cubic-dev-ai]

All reported issues were addressed across 3 files (changes from recent commits).

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

All reported issues were addressed

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

All reported issues were addressed across 3 files (changes from recent commits).

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Fix all with cubic | Re-trigger cubic</sub>