anotifier: the Claude Code notifier for desktop, phone and Slack.

Searching for a Claude Code notifier turns up five kinds of tool: Claude Code's own notifications, companion apps, native macOS tools, VS Code extensions, and hooks. Each suits someone. Facts about other projects come from their own pages, linked below, and may have changed since 2026-10-15.

Pick by what you need to reach. One platform and one editor: a companion app or an extension may fit better. More than one OS, your phone or a chat channel, or more than one agent: a hook-based tool like anotifier covers them with one install. For how each hook option works, the Claude Code notification guide walks through them.

Claude Code runs hooks: shell commands registered in a settings file. Your user settings file is ~/.claude/settings.json (%USERPROFILE%\.claude\settings.json on Windows), and it applies to every project. Stop fires when Claude finishes a turn; Notification fires when it is waiting for permission or input. This pair shows a macOS banner for both:

{
  "hooks": {
    "Stop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "osascript -e 'display notification \"Claude finished\" with title \"Claude Code\"'"
          }
        ]
      }
    ],
    "Notification": [
      {
        "matcher": "",
        "hooks": [
          {
            "type": "command",
            "command": "osascript -e 'display notification \"Claude needs your input\" with title \"Claude Code\"'"
          }
        ]
      }
    ]
  }
}

Swap in the command for your OS. Inside the JSON, write each double quote as \" (the macOS and Windows commands both contain some). The Windows one is the dialog box from Claude Code's own docs; it stays on screen until you click OK, so use it for Notification rather than Stop. On macOS, osascript notifications are attributed to Script Editor; if Script Editor is not allowed to send notifications, the command fails silently.

# macOS
osascript -e 'display notification "Claude finished" with title "Claude Code"'

# Linux (needs a notification daemon; on Debian/Ubuntu libnotify-bin provides notify-send)
notify-send 'Claude Code' 'Claude finished'

# Windows (PowerShell)
powershell.exe -Command "[System.Reflection.Assembly]::LoadWithPartialName('System.Windows.Forms'); [System.Windows.Forms.MessageBox]::Show('Claude finished', 'Claude Code')"

• If the file already has a hooks key, add Stop and Notification next to the events already there rather than replacing the object. Type /hooks in Claude Code to check they registered.
• The Notification hook with an empty matcher runs for every notification type. To alert only for approvals, set "matcher": "permission_prompt"; see the permission notifications guide.
• Claude Code's documentation names ~/.claude/settings.json as the user settings file. If you set CLAUDE_CONFIG_DIR, Claude Code reads its settings from that folder instead.

What anotifier adds on top of this:

• A useful message. The title names the project (my-app · Claude Code) and the toast carries the start of Claude's last reply or its question, instead of fixed text.
• Phone and chat in the same hook, sent to every channel in parallel with a short timeout each, so a dead channel never holds Claude up.
• Priorities. Permission prompts and questions go out urgent; the idle reminder and informational types go out at normal priority.
• No double alerts. A repeat of the same event within 1.5 seconds is dropped.
• Snooze and quiet hours, an uninstall that removes only its own entries, and a backup of settings.json before it edits.
• Usage and context warnings through the statusline, covered below.

Setup only wires Claude Code when ~/.claude/settings.json already exists (on Windows, in the .claude folder of your user profile), so start Claude Code once first. It copies the file to ~/.anotifier/backups/, then adds this entry under Stop:

{
  "hooks": {
    "Stop": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "node \"/path/to/anotifier/src/notify.mjs\" --source claude",
            "timeout": 10
          }
        ],
        "_managed_by": "anotifier",
        "matcher": ""
      }
    ]
  }
}

Notification gets an identical entry. The path points at the anotifier package on your machine, which is the npx cache if you ran it with npx. _managed_by is how status and uninstall tell anotifier's entries from your own hooks. Claude Code gives the hook 10 seconds; anotifier sends to every channel in parallel, each with a shorter timeout, and always exits cleanly, so a dead channel never holds Claude up.

From 1.5.0, setup also routes the statusline through anotifier, which is how usage-limit warnings (1.5.0) and context warnings (1.6.0) work. It adds this statusLine entry; the --wrap-b64 part appears only if you already had a statusline, and carries your old command, base64-encoded, so it keeps printing:

{
  "statusLine": {
    "type": "command",
    "command": "node \"…/anotifier/src/statusline.mjs\" --wrap-b64 <your old command, base64>"
  }
}

Installed as a plugin instead? The plugin registers the same two hooks from its own hooks.json, and /anotifier:setup wires your other agents. Plugin hooks don't show in anotifier status and anotifier uninstall doesn't remove them; remove the plugin from Claude Code's /plugin menu instead.

• The ding comes from Claude Code itself. The hook hands Claude Code a bell to print in its own terminal (Claude Code 2.1.141 or newer), so no extra process has to find the right tab. To change the sound or turn it off, see the Claude Code notification sound guide.
• Only types that wait on you are urgent. Permission prompts and questions go out urgent. Informational types such as agent_completed arrive at normal priority, and echoes of something you just did, such as auth_success, are dropped. The hook has no matcher; the permission notifications guide shows how to filter by type in a hook you write yourself.
• Needs-input alerts stand out. They go out urgent, which raises the priority on ntfy and Linux notify-send, and toasts play a different sound than for a finished task.
• Click-to-focus is Windows-only. Clicking the toast brings forward the window whose title contains the project folder. macOS, Linux and WSL toasts have no click action.
• Snooze and quiet hours silence every channel, the bell included: npx anotifier@latest snooze 2h.

Version 1.5.0 and later warns you before Claude Code stops at a usage limit. One notification goes out as your 5-hour or weekly usage crosses 70%, 85% and 95%, at default, high and urgent priority. It names the chat, the window and when it resets. Usage belongs to your account, so each threshold warns once per window across all your chats.

• Claude.ai Pro and Max only. Claude Code reports usage only to those plans, and only to the statusline command, never to hooks.
• Setup routes your statusline through anotifier. anotifier setup points statusLine at src/statusline.mjs. A statusline you already have is wrapped and still prints your line; anotifier uninstall restores it exactly.
• Toast, ntfy and webhook, never the bell. Snooze and quiet hours hold warnings back.
• Configurable. usageAlerts: { "enabled": true, "thresholds": [70, 85, 95] } in ~/.anotifier/config.json.

The usage limit notifications guide has the statusline script to write yourself and the full details.

Version 1.6.0 and later sends one notification per chat when its context passes 85% of the auto-compact window, so you can wrap up or run /compact before Claude Code compacts on its own. It names the chat and the size, such as 174K of 200K tokens, at high priority.

• Once per chat. A chat never warns twice, even after /compact. /clear starts a new session, which can warn again.
• Measured against the auto-compact window, from CLAUDE_CODE_AUTO_COMPACT_WINDOW or your /autocompact setting, else the model's window. Skipped when auto-compact is off. The --autocompact flag and managed settings are invisible to a statusline, so the warning can be late or missing with them.
• No subscription needed. It works on API-key and proxy sessions too, through the same statusline wiring as usage warnings.
• Toast, ntfy and webhook, never the bell. Snooze and quiet hours hold it back.
• Configurable. contextAlerts: { "enabled": true, "threshold": 85 } in ~/.anotifier/config.json.

The auto-compact warning guide explains how the window is resolved and has a statusline script to write yourself.

npx anotifier@latest test       # fire a test alert on every channel
npx anotifier@latest status     # show which agents and channels are set up
npx anotifier@latest doctor     # diagnose a channel that stays quiet
npx anotifier@latest uninstall  # remove the hooks setup added

test sends straight to your channels without going through Claude Code, so it proves the channels work. uninstall removes only the entries anotifier added and restores your statusline (1.5.0 and later), and setup keeps a copy of every file it changes in ~/.anotifier/backups/.