Browser Control

Status: dark, off by default. Requires Machines, the Browser Control platform gate, your organization's signed-in owner enabling the mode, and the person at the browser enabling it and approving each session. Nothing here enables it.

An agent can use sites you are already signed in to in your own Chrome profile: open tabs, search and read pages, click, type, navigate, inspect a DOM tree, take masked screenshots, and collect a short screenshot sequence. No extra browser subscription is needed. All page content is treated as untrusted data.

The person chooses exact sites and a named control lease. A session defaults to 30 minutes, cannot exceed four hours, and stops after five idle minutes. Banking, payment, password-manager, webmail and sign-in provider pages are blocked. Pages containing password/OTP/payment controls or frames are refused as a whole. Passwords, form values, cookies and storage are never returned. Frame control is unavailable.

Stop Browser Control in extension Options or on the page banner detaches agents immediately. The shortcut is Ctrl+Shift+K (Command+Shift+K on Mac); Chrome may need you to choose a different shortcut if another extension already uses it. Every action is audited locally and through Machines. The toolbar badge and page banner show agent access.

Diagnostics return console levels/argument types and network method/host/status/type rather than raw messages, headers or bodies that could contain secrets. Constrained JavaScript is a finite read-only set over sanitized page data. File upload requires an org-approved attachment id also approved in the local session. Batch actions recheck consent at each step.

• Runbook: owner approval and enrollment.

Setup: the extension guide. Contract and limitations: Browser Control design.

API: OpenAPI. MCP: tool reference and MCP tools.