Skip to main content
Some services have no OAuth, no MCP server, and no API. The agent can still use them, by signing in through the shared browser and holding the session. This is the fallback that makes “anything you can do, your agent can do” true rather than aspirational.

​
How it works

1

The agent opens the site

It navigates to the sign-in page in its own browser and stops there.
2

You sign in yourself

The agent sends you a link to watch and control that browser. You enter the password, the code, or whatever the login needs. The credential never passes through the agent or the conversation.
3

It carries on

The session is held in the browser, and the agent continues the task from where it stopped.
For services that support neither MCP, OAuth, nor API keys, the agent can use the shared browser. You log in yourself in the browser. The session is captured so the agent can keep working in that site.

​
When to use it

Reach for this when there is genuinely no programmatic option, or when a login must be yours: an SSO flow, a CAPTCHA, a two-factor code. Prefer OAuth, MCP, or an API key whenever the service offers one, since a held session is more fragile than a token.

Browser use

What the agent’s browser can do, and how the shared view works.

Catalog

Check whether the service has a cleaner method first.