Claude Code notifications when done or waiting for input.

In Ghostty, Kitty, and iTerm2, Claude Code already sends a desktop notification when it finishes or pauses for permission, and it reaches your local machine even over SSH. In other terminals, switch the notification channel to the terminal bell. Most guides still tell you to run `claude config set --global preferredNotifChannel terminal_bell`, which now fails with `error: unknown option '--global'`: the config subcommand was removed, so set it in settings.json instead.

Or run the one-liner below, which adds a Notification hook that rings the bell and raises the desktop alert itself, so it works whatever channel you are on. Either way it stays on this machine and cannot reach you once you step away.

Claude Code fires a Notification event when it needs your approval (permission_prompt) or is done and waiting for your next prompt (idle_prompt). A hook in .claude/settings.json turns that into any command you want, like a macOS notification via osascript or notify-send on Linux.

Add a matcher such as permission_prompt to only fire for specific types. This is the mechanism most DIY tutorials skip: no prompt hacks in CLAUDE.md, just a documented event.

The Stop event fires when Claude finishes responding. Point a Stop hook at ntfy.sh and your phone gets a push for free. The trade-offs: it is one-way, so when Claude is blocked on a question you still have to walk back to the terminal to answer, and public ntfy topics are readable by anyone who guesses the name unless you reserve one or self-host.

Claude Code has its own mobile push, and it only works while Remote Control is on. Start the session with claude remote-control, run /remote-control in one that is already open, or turn on Enable Remote Control for all sessions in /config.

Install the Claude app on your phone, sign in with the same account, and allow notifications. Then run /config and switch on Push when actions required for permission prompts and questions, Push when Claude decides for pushes Claude sends on its own, typically when a long task finishes, or both.

You can also ask for one in your prompt, such as notify me when the tests finish. Claude picks the moment, there is no per-event setting beyond those two switches, and it skips the push while you are typing in or focused on the terminal.

You can answer the prompt from the app. It needs a Pro, Max, Team or Enterprise plan signed in through claude.ai, and on Team and Enterprise an Owner has to turn Remote Control on first. API keys, Amazon Bedrock, Google Cloud, Microsoft Foundry and a custom ANTHROPIC_BASE_URL are not supported, and your computer has to stay on.

One command installs the hooks and connects Claude Code over MCP. You get a push when Claude finishes, and when it is blocked on a permission or a question, the notification carries the question itself. Tap yes or no from your lock screen and Claude continues.

At your desk, the Pushary Mac app shows the same question in the notch. Permission policies decide which tools need your approval, and every decision lands in an audit trail. This is the difference from every free option above: they tell you Claude is waiting, Pushary lets you unblock it.

Claude's own Remote Control covers Claude sessions signed in with a claude.ai account. Pushary works however Claude Code logs in, including API keys, Bedrock and Vertex, and puts Codex, Cursor and Gemini CLI in the same app.