Muster tells you the moment Claude Code, Codex or Cursor needs you, and lets you answer with one keystroke, from whatever you're doing.
Two agents asked. Two keystrokes, and you never left your editor. ⌥↩ allows, ⌥⌫ denies, on any screen, notch or not. For Claude Code and Codex, once you turn it on. If Muster quits, your agent just asks the usual way.
Press ⌥↓ and the alert unfolds: the agent's reason for asking, the whole command, the file and size of an edit, or every option of a question with its note.
Pick an option with ⌥1, ⌥2 or ⌥3, or fold it away with ⌥↑. The alert always shows which keys it takes.
Hold ⌥ and tap Tab. Every session is there, the one waiting on you first. Let go and you're in its window, or the terminal it runs in.
Every session from the last three hours, in one list. Start typing to narrow it down.
Each one shows its tokens and what they would cost at API list prices. On a Max or Pro plan you don't pay that, but it tells you which sessions are the expensive ones.
Muster is a native AppKit app and a 2 MB download. It uses the system's own glass and follows your light or dark appearance.

Claude's app, the Codex thread, Cursor's Agents window, or the terminal a CLI session runs in.
Claude Code sessions that the desktop app runs over SSH show up next to your local ones.
Notifications carry Allow and Deny, or the options to a question.
Change ⌥Tab and ⌃⌥Space. Muster tells you if another app already uses one.
muster://next opens the session that needs you most. muster://search opens search.
A finished session stays marked until you open it.
Muster takes ⌥↩, ⌥⌫ and the arrows only while something on screen needs them, and lets go while you type elsewhere.
A session stopped by a usage limit moves out of your way and shows when the limit resets.
After an update, Muster shows what's new a single time. Every release is in the changelog.
Muster reads the session files your agents already write. An agent's question reaches Muster over localhost, and your answer goes back the same way. There's no account and no analytics, and the update check stays off until you turn it on.
Muster reads their session files.
Questions and answers stay on 127.0.0.1.
An app that can approve what your agents run should be one you can read. Muster is MIT licensed, and every release passes 213 end-to-end tests that drive the real app before it ships.
Read the code on GitHubMuster counts the sessions waiting on you in the menu bar and shows an alert at the top of your screen. To compare it with Claude Code's own options, read getting notified when Claude Code needs you.
Yes. Press ⌥↓ on the alert to see the agent's reason, the whole command, or every option of a question with its note, then answer with ⌥↩ or ⌥1 to ⌥9.
No. Muster takes ⌥↩, ⌥⌫, the arrows and the digits only while a prompt or the switcher is on screen, and lets go of them while you type in another app.
Yes. Every Claude Code, Codex and Cursor session from the last three hours is in one list, with the one that needs you first.
Yes. Muster is open source under the MIT License, with no account.
No. It reads the session files your agents write and talks to them over 127.0.0.1. The only network request is an update check you can turn on.
The agent falls back to its own prompt, the same as without Muster.
Not yet. Cursor doesn't let another app answer for it, so Muster shows you which chat is waiting and opens it in Cursor.
macOS 14 or later, on Apple silicon and Intel.
Every release is in the changelog. Muster also shows what changed after each update.