Claude Code notch app for macOS — the complete guide

Why a Claude Code notch app exists

Claude Code is a long-running, interactive agent that pauses roughly 8 to 15 times an hour to ask for permission — to run a bash command, write a file, call a tool. If you sit in the terminal and watch, you approve instantly and no time is lost. The moment you context-switch to another app, every approval Claude waits on becomes idle time. A single session leaks 30–50 minutes a day this way. Run three or more sessions in parallel and you can lose half your compute to missed approvals.

A Claude Code notch app solves this specific problem. It hooks into Claude Code at the source — events fire inside the running process — and surfaces them in the MacBook notch, which is visible in every Space, on every full-screen app, in every workflow. Your peripheral vision is already pointed at the top of the screen. You notice a dot change colour, you tap or press a shortcut, you go back to what you were doing. No alt-tab, no terminal hunting, no cognitive reset.

How a Claude Code notch app works under the hood

All current Claude Code notch apps share the same plumbing:

1. Hook registration. On install, the app writes shell scripts into your Claude Code hook directory. When Claude Code reaches a state transition — session start, tool call requested, approval needed, completion — it executes the hook script, which sends a JSON payload over a Unix domain socket to the notch app.
2. Notch rendering. The notch app receives the payload, decides which session it belongs to, and updates the notch UI. State is rendered ambiently (breathing dots, status colours) until something needs your attention, at which point the notch expands.
3. Approval routing. When you click Approve or press a keyboard shortcut in the notch, the app needs to deliver the keystroke back to the right terminal pane. The implementations differ here — some use macOS Accessibility to send keys to the focused window, others identify the terminal by PID and route via the terminal's scripting bridge.
4. tmux + multiplexer support. Inside tmux the operating system sees one process per terminal — multiple Claude Code sessions live as children of that single PID. Apps with deep tmux support capture the actual PID at the hook source and route approvals via tmux send-keys to the correct pane. How Claude Pulse implements this →

None of these notch apps rely on screen scraping, OCR, or heuristics about what's on screen. The hook is a stable interface that Anthropic ships with Claude Code itself, so state changes show up in the notch within a second of happening in the agent.

What to evaluate when choosing a Claude Code notch app

1. Safety model on batch approval. If the app lets you approve all queued tool calls at once, ask whether destructive commands — rm -rf, DROP TABLE, git push --force — are filtered out of the batch automatically. Many apps don't publish a clear answer.
2. Focus-awareness. When a session you're already looking at needs approval, an aggressive notch expansion steals attention from work you're doing in the terminal. Apps that capture session identity at the hook source (inside the terminal tab) can bounce the notch instead of expanding it.
3. Live plan usage. Your 5-hour rolling window and weekly Anthropic limit govern how aggressively you can run sessions. Apps that read the Claude CLI's existing Keychain credentials can show these live, in the notch footer, without a second OAuth flow.
4. Multi-session ergonomics. A keyboard shortcut to cycle through sessions, a clear count, breathing-dot status per session, context window fill bars — these compound when you run three or more sessions simultaneously.
5. Terminal coverage. Ghostty, iTerm2, Terminal.app, WezTerm, Kitty, Warp, Alacritty, Hyper, Tabby, tmux — different apps cover different subsets. Pick one that covers the terminal you actually use.
6. Privacy and locality. Verify the app runs entirely locally and doesn't ship session contents to a cloud service. The Claude Code hook payload contains the tool name and arguments — sometimes file paths or partial code — and you don't want that leaving your Mac.
7. Pricing model. Most current Claude Code notch apps are free; one is paid ($20 one-time). The free options are competitive — start free, pay only if a specific paid feature you need is missing.

The seven Claude Code notch apps available today

As of May 2026, these are the notable Claude Code notch and notch-adjacent apps for macOS. For an honest side-by-side table with capability rows and "?" markers where claims aren't publicly verified, see the full alternatives comparison.

• Claude Pulse — what we built →

  Free. Safety-aware batch approval, live Anthropic usage, focus-aware notch, 9 terminals + tmux.

• Notchi

  Free / paid tiers. Covers both Claude Code and OpenAI Codex hooks in one notch.

• Vibe Island

  $20 one-time. Polished notch UI, approve from any app, multi-session.

• Claude Island / Vibe Notch

  Free, open source. Minimal, always-present session manager focused on context-switch-free notifications.

• Notch Pilot

  Free, open source. Live companion that reacts to Claude state and intercepts permission prompts.

• AgentNotch

  Free, open source. Menu bar + notch presence, watches Claude Code and Codex thinking.

• c9watch / Claude Status

  Free, open source. Menu bar with native WidgetKit widgets; discovers sessions at the OS process level.

How to install a Claude Code notch app

Installation is similar across all seven apps. Using Claude Pulse as the example:

1. Download the DMG from claudepulse.app/download. The DMG is notarized by Apple, so Gatekeeper accepts it without any workaround.
2. Open the DMG and drag the app into Applications.
3. Launch the app. On first run, grant macOS Accessibility permission (so the app can route approval keystrokes back to the terminal) and Automation permission for the terminal you use (so it can identify which window a session belongs to).
4. Open any Claude Code session. The notch will pick it up automatically within a second. No restart, no config.

If you switch to a different Claude Code notch app later, uninstalling is a drag-to-trash; the hook scripts get removed with the app bundle. There is no global lock — multiple notch apps can technically coexist, though only the one your hooks point at will receive Claude Code events.

Frequently asked questions

Does a Claude Code notch app slow down Claude Code?

No. The hook is a fire-and-forget shell exec when Claude Code reaches a state transition. Hook execution is parallel with Claude's own work and the notch app only processes the payload after the hook returns. Measured overhead is in single-digit milliseconds.

Will a notch app work with Claude Code in Docker, Lima, or distrobox?

No. The hook socket lives on the host filesystem and isn't reachable from inside a container namespace. Sessions launched inside a container won't appear in any current notch app. Native host sessions and tmux are fully supported.

Can a Claude Code notch app see my conversation contents?

It can see the hook payload, which contains the tool name, arguments, and a small status string. It does not see the full conversation transcript or the prompt you sent. Whether that data leaves your Mac depends on the specific app — see each app's privacy policy. Claude Pulse keeps everything local.

What happens when I'm using a MacBook without a notch?

Some apps fall back to a floating window near the top of the screen. The design is optimized for the hardware notch on 2021+ MacBook Pro and 2022+ MacBook Air, so older MacBooks get a usable but less elegant experience.