Claude Code in your MacBook notch

The problem is the waiting, not the working

Claude Code does most of its work unattended. The part that costs you the day is the pause: it hits a command it needs permission for, stops, and waits in a terminal tab you are not looking at. You find out minutes later, when you go check.

CrewTower turns that pause into something that reaches you. The session's card comes forward in the notch with the exact command on it, and you answer from there.

Approve without touching the terminal

When Claude Code asks to run something, its hook blocks on a local Unix socket while it waits for an answer. CrewTower replies over that same socket, so the prompt resolves in place - your terminal never comes forward and nothing is typed into it.

A destructive shell command is flagged on the card before you press anything, so a fast Allow stays a safe Allow. And there are three ways to say yes: the button allows once; the menu beside it allows every permission request for the rest of that session, a grant CrewTower keeps itself rather than a rule in Claude Code's settings; and, when Claude Code proposes a rule for the request, Always allow has Claude Code add it to the project's local settings, the same file you would have edited by hand.

Answer its questions and its plans, too

Permission requests are not the only thing Claude Code stops for. When it asks you to choose between options, those options render as buttons and your pick goes back verbatim in the shape Claude Code expects. When it puts up a plan, the steps are on the card and you approve or send it back from there.

A tool with no bespoke card falls back to a generic one carrying the flattened parameter list, so the notch still shows what the call will do and still offers the same answer the terminal offers.

Tell Claude what's next

When Claude finishes and you already know the next step, you no longer have to find its tab to say it. Type it in the card's Tell Claude what to do next field and press Send, and it goes into that same session as your next prompt. If Claude is still working, it runs when the turn ends.

The card's header has Stop, which ends the running turn after asking once, and the context percentage opens the same breakdown /context draws, with Compact and Compact with a focus under it. It works in the terminal and in Claude Desktop's Code tab.

New to the idea of a Claude Code plugin that draws on the card and sends prompts? How Claude Code mods work explains what a mod is, how it differs from a hook and what it is trusted with, and the best Claude Code mods lists eighteen others with public code.

Land on the exact tab

When you do want to be in the terminal, Jump takes you to that session's own place, not just the application: the exact tab in Terminal, iTerm2 (down to the split) and Ghostty; the pane in WezTerm, tmux and herdr, and in kitty with its remote control on; Warp's tab through the focus link Warp provides; and the window in Alacritty, which has no tabs. How it resolves each of them is written up in full.

macOS Accessibility permission is used for Jump in the terminals that need it, such as Ghostty and Alacritty, and to tell whether you are already looking at a session. Approving from the notch does not need it.

Setup is one launch

Claude Code reads its hooks from its own settings. CrewTower writes that configuration the first time it runs and removes it again when you delete the app - there is no file for you to edit and nothing left behind afterwards.

The next time Claude Code stops for you, the card is in the notch.

It stays on your Mac

Sessions are read from what Claude Code already writes locally, and decisions travel over a local socket. The optional plugin behind Tell Claude uses a local socket too, and makes no network request. There are no accounts, nothing to log into, and no telemetry in the app at all. Transcripts, prompts and diffs never leave the machine.

What it does not do

CrewTower shows how much of your subscription window is left, read from Anthropic's own usage endpoint through your existing Claude Code login, plus how much context each session has left. What it does not do is keep a token or cost history: beyond the current session's running cost, which a card can show with the beta Tell Claude plugin on, a breakdown of what Claude Code has spent by day, session or model is a different question and a different tool. It is not a terminal, not an IDE, and not a way to run agents - it watches and answers the ones you started yourself.

Questions

Does CrewTower need Claude Code hooks configured by hand?

No. CrewTower writes the hook configuration on first launch and cleans it up when you remove the app. There is no settings file to edit yourself.

Does approving from the notch send keystrokes to my terminal?

No. Claude Code's hook blocks on a local Unix socket while it waits for a decision, and CrewTower answers over that same socket. The prompt resolves in place. There is no synthetic-keystroke path anywhere in the app.

Can I send Claude Code its next prompt from the notch?

Yes, in beta. Turn on Tell Claude from the offer on a Claude card or in Settings > Integrations, then start a new session or type /reload-plugins in an open one. Each Claude card can then send its session a prompt, stop the turn and compact it. It needs Claude Code 2.1.287 or newer, 2.1.286 in Claude Desktop, and turning it off uninstalls it.

Does it work with more than one session at a time?

Yes. Every running session gets its own card, and the ones waiting on you sort to the front.

Does CrewTower show my Claude Code usage?

It shows how much of your subscription window is left: the 5 hour session window and the weekly window, as a percentage used with a countdown to the reset. That reading comes from Anthropic's own usage endpoint using your existing Claude Code login, so it is the provider's own number rather than an estimate parsed from local files. The meter is on by default and can be turned off in settings, and it also shows how much context each session has left. It keeps no token or cost history by day, session or model. With the beta Tell Claude plugin turned on, a card can show what the current session has cost so far: Claude Code's own /cost figure, at API prices. For a history, ccusage is the tool most people use.

Does CrewTower send my Claude Code sessions anywhere?

No. Transcripts, prompts and diffs never leave the Mac, and there are no accounts, no cloud sync and no analytics in the app. The only network requests the app makes are the optional usage meter, license activation and update checks, and the privacy page lists all three.

Does it work with agents other than Claude Code?

Yes - Codex, Cursor, Gemini CLI, Qwen Code and opencode are all spoken natively. The twenty other agents CrewTower recognises get a card you can jump from, and the agents page lists all twenty-six with what it does for each.