Get a phone alert when your AI agent needs you.

When the agent runs on your computer, every route here needs that computer to stay on.

Anthropic's phone route for Claude Code is Remote Control: while it is active, Claude can send push notifications to your phone through the Claude mobile app.

1. Install the Claude mobile app on your phone and sign in.
2. In your Claude Code session, type /remote-control to connect it.
3. Type /config and turn on Push when actions required, Push when Claude decides, or both. Both are off by default.

Anthropic says Claude decides when to push: it typically sends one when a long-running task finishes or when it needs a decision from you to continue. Beyond the two switches, there is no per-event setting.

It needs a Pro, Max, Team or Enterprise plan; API keys are not supported. Your computer has to stay on and the claude process has to keep running. Pushes are skipped while you are typing in, or focused on, the terminal.

Anthropic's Cowork article says that when Claude finishes a task or needs your input, you get a notification on your phone. On Pro and Max plans, new Cowork tasks run in the cloud, so they keep going after you close your laptop.

Dispatch is a separate route. If you already have it on an eligible Pro or Max plan, you can hand tasks to your computer from your phone. Anthropic says: "You'll get a push notification on your phone when a task is done or when Claude needs your go-ahead."

• Dispatch is closed to new users. It stays available only to some existing Pro or Max users, and not on Team or Enterprise plans.
• For tasks that run on your computer, keep the Claude Desktop app open and the computer awake.
• The task states Awaiting input and Awaiting answer mean it waits for you.

What the Mac shows: Know when a Claude Cowork task is done or needs you.

Since 9 July 2026, Codex is part of the ChatGPT desktop app. Its Settings > Notifications sets when turn-completion alerts appear (never, only while ChatGPT is in the background, or always) and has separate switches for permission and question alerts. These show on the Mac.

ChatGPT on the web has its own Settings > Notifications, with channels such as push, email or SMS depending on the category and your account. OpenAI's docs don't say that these send local Codex turn completions, permissions or questions to your phone. For long local work, turn on Prevent sleep while running so the Mac stays awake.

The Codex CLI.

We found no phone setting for the Codex CLI in OpenAI's docs, and its notify setting never fires for approvals. A hook can do it: PermissionRequest runs when Codex is about to ask for approval, and Stop when a turn ends. Codex skips new or changed hooks until you trust them in /hooks. More in See when Codex needs your approval, or finishes.

A hook is a command that an agent runs at set moments, such as when it finishes a turn or waits for your OK. ntfy.sh is a public service that passes short messages to the ntfy app: the hook sends a line to a topic, and the app, subscribed to that topic, shows it.

1. Install the ntfy app on your phone and subscribe to a topic with a long, random name.
2. Add the snippet to ~/.claude/settings.json (all your projects) and put your topic in place of REPLACE-WITH-A-LONG-RANDOM-TOPIC in both places.
3. Edits are normally picked up on their own; if not, restart the session.

In ~/.claude/settings.json

{
  "hooks": {
    "Stop": [
      { "hooks": [ { "type": "command", "command": "/usr/bin/curl -s -m 10 -H 'Title: Claude Code' -d 'Claude Code finished' https://ntfy.sh/REPLACE-WITH-A-LONG-RANDOM-TOPIC >/dev/null 2>&1", "async": true } ] }
    ],
    "Notification": [
      { "matcher": "permission_prompt", "hooks": [ { "type": "command", "command": "/usr/bin/curl -s -m 10 -H 'Title: Claude Code' -H 'Priority: 4' -d 'Claude Code needs you' https://ntfy.sh/REPLACE-WITH-A-LONG-RANDOM-TOPIC >/dev/null 2>&1", "async": true } ] }
    ]
  }
}

What each part does.

• Finished. Stop fires whenever Claude finishes responding, not only at the end of a whole task. The public ntfy.sh server allows 250 messages a day.
• Waiting for you. permission_prompt fires once a permission prompt has waited about six seconds while you seem to be away; each keystroke in the terminal puts it off. Priority: 4 marks the message high in ntfy.
• No waiting. "async": true lets Claude Code carry on meanwhile.

If nothing arrives.

• Trust the folder. Claude Code holds back hooks, even your own, until you accept that folder's workspace trust dialog.
• Work Macs. An organization can allow only its own managed hooks, and then personal hooks don't run.
• If you also use Cowork Island. Keep each of your hooks in its own entry, as in the snippet, and keep your own scripts outside ~/.perch/. Cowork Island treats commands containing /.perch/ as its own and may replace their whole entry.

For Codex CLI or Gemini CLI, the notify hook generator builds the matching snippet with a random topic, names its file and can merge it into a pasted copy, all in your browser.

• Claude Code docs: Remote Control
• Claude Code docs: Settings reference
• Claude Code docs: Hooks reference
• Claude Code docs: Hooks guide
• Claude Code docs: Settings
• Claude Code docs: Desktop app
• Claude Help Center: Dispatch in Cowork
• Claude Help Center: Cowork on web, desktop and mobile
• Anthropic: Dispatch task states
• ChatGPT docs: Notifications
• ntfy: Subscribe from your phone
• Codex changelog
• Codex docs: Long-running work
• Codex docs: Advanced configuration
• Codex docs: Hooks
• Gemini CLI docs: Hooks
• ntfy docs: Publishing

Sources checked 18 October 2026. This guide is written by Cowork Island's maker. For corrections, email support@coworkisland.com or open an issue on GitHub.