AI Automation

AI automation lets an AI assistant work in Studio on your behalf. It can write and edit scripts, run them in local simulation, and look at what they do. It uses the Model Context Protocol (MCP), which Claude Code, Codex and the Claude desktop app support.

You stay in charge throughout. The assistant can only start after you approve it in Studio, and only one assistant can work in Studio at a time. You can take back control at any moment.

AI automation is available on macOS. If Settings → AI Automation is missing, your version of Studio doesn’t include it yet.

Turn it on

Open Settings → AI Automation.
Turn on Enable AI automation.
Under Connect an AI client, follow the steps for your assistant.

The connection runs inside Studio itself, so there is nothing else to install. Studio never changes another app’s settings on its own; it gives you setup to copy, or asks the other app to confirm.

Connect your assistant

Claude Code: copy the command and run it in Terminal. Claude Code can then use Studio in every project.
Codex: copy the command and run it in Terminal. The Codex app and the Codex command line share this setup.
Claude desktop app: choose Add to Claude…. Claude explains what it is adding and asks you to confirm. If you prefer, copy the configuration entry into Claude’s configuration file instead (Show Configuration File finds it), then quit and reopen Claude.

Other assistants that support local MCP servers may also work with the configuration entry, but only the three above are supported.

Keep Studio in your Applications folder. If you move or rename it, Settings shows the updated setup; copy it into each assistant again. After you update Studio, restart your assistant’s connection to Studio, or start a new session, so it picks up the new version.

Approve the assistant

When an assistant asks to work in Studio, Studio asks you first. Choose Allow once, Allow until Studio quits or Don’t allow. The name shown is the one the assistant gives itself; Studio can’t verify it.

Studio may also ask before some steps:

Network or storage access. A script that wants to reach the network or keep data between runs needs your permission.
Bluetooth. Scanning for, connecting to or writing to real devices, and advertising over your Mac’s radio, each need your permission. A device’s name is whatever the device says it is.

The assistant can’t approve anything for you.

While the assistant works

The toolbar shows AI in control, and editing and run controls are paused, so the two of you never change the same thing at once. Studio won’t hand over control while you have unsaved changes or a session running; save or stop first, then let the assistant ask again.

Work the assistant saves stays in your script library after it finishes. Scripts it created or last changed are labelled, and a script it deletes goes to Script Trash, where you can restore it.

Take back control

Choose Take back control in the toolbar or in Settings at any time. Studio stops the assistant’s work, returns your controls, and pauses AI automation, so the assistant can’t simply ask again.

When you’re ready, choose Allow AI automation again.

If Studio can’t confirm that everything stopped, it keeps your controls locked and shows Retry Stop rather than guessing.

Trusted integrations

If you work with the same assistant often, a trusted integration saves you from approving every step. Under Trusted integrations, choose Add trusted integration…, give it a name, and choose:

how long it lasts: Allow once, Until Studio quits or Until revoked;
whether its scripts may use the network or keep data between runs.

Studio then gives you setup for that one assistant. It contains a private key that identifies the integration:

Add it only to the assistant you meant it for.
Never paste it into a chat, an issue, or anywhere someone else could read it.
Running a setup command saves it in your Terminal history. Save private configuration… avoids that.

If the setup is lost or shared by mistake, choose Revoke and add a new integration.

Trust covers the scripts in your current library. Studio still asks about anything new, such as Bluetooth access to another device. If you choose Remember, it won’t ask again for that exact device or target. In Settings you can review what each integration may do, remove permissions, or revoke it. If you take back control from a trusted integration, it stays paused until you choose Allow automation again for it.

Current limits

AI automation runs scripts in local simulation unless you approve a run on your Mac’s Bluetooth radio.
Each step on real Bluetooth devices needs its own approval.
The assistant never buys anything, opens purchase screens, or starts a trial day.
AI automation is still being tested with each supported assistant and with real devices, so some versions of Studio ship with it turned off.