Why let an AI agent use your own signed-in Chrome?

There are two common ways to give an AI agent a browser. You can start a fresh automated browser for it, or you can let it use the Chrome you already have. Beacon takes the second route. This post explains why, and what you give up.

A fresh automated browser starts empty. It has no cookies, no saved sessions and none of your devices' trust. Anything behind a login means the agent has to sign in, and signing in is exactly where automation tends to struggle: passwords, two-factor prompts, "is this you?" emails and sign-in walls designed to slow down anything that is not a person.

You can work around some of it, but the workarounds are the part nobody enjoys. For tasks in accounts you already use, you spend more effort on getting in than on the task.

Your own Chrome already has your sessions, your 2FA and your cookies. You signed in once, the way you normally do. When the agent works in a tab inside that browser, there is no re-login and no sign-in wall to fight. The task starts at the useful part.

It also means the agent sees the same sites you do, with the same accounts, which is the point for admin tools, dashboards, inboxes and anything else that lives behind your login.

Using your own browser is only reasonable if the agent has boundaries. Each agent works in its own lane, which is its own tab group and its own tabs, and it cannot act on another lane's tab or on yours. Irreversible steps such as sending or paying are flagged so the agent can ask first. In agent mode your own tab is left untouched, and a tab only comes forward when the agent needs a human.

What the agent does inside its lane still depends on the instructions you give it. Read is it safe to give an AI agent your signed-in browser for the full picture.

The agent can reach what you can reach, inside its own tab group, so you need to think about what you ask it to do. With a fresh browser there is nothing for the agent to misuse, because there is nothing in it. With your own Chrome you trade that emptiness for usefulness, and the lanes and flagged steps are what keep the trade reasonable.

Any MCP client works, including Claude Code and Cursor. Beacon is exposed as a remote MCP server over HTTP with a Bearer key, so you paste one config entry into your client's settings, or use the one-line command for Claude Code. The setup guide has both.

The agent's commands, and what the page sends back, pass through the hosted relay to the extension. Treat the relay like any hosted service. If you would not send a given page's contents through a hosted service, do not point an agent at it. The privacy policy describes what Beacon collects.

A fresh browser is the right tool when you want a clean, repeatable environment, for example when running tests, or when the work has nothing to do with your accounts. Beacon is for the other case: a task that needs to happen in the accounts you are already signed into.

If you are comparing approaches, Beacon and Playwright for logged-in sessions lays out the difference without picking a side.