I make Tapproval, which is the third option here. So I'm biased, and you should read this knowing that. Checked on 19 November 2026.
Claude Code stops a lot. It wants to run a command, or it has a question, or it's finished a plan and wants a yes. If I'm at the desk that's fine. If I've gone to make coffee, it sits there until I'm back, and a twenty-minute job takes an hour.
I wear an Apple Watch, so the obvious fix is to have the watch tell me. I've tried three ways of doing that. Two of them are free and you may not need anything more.
The built-in one: Remote Control
Claude Code can already push to the Claude app on your phone, and the watch mirrors the phone. Setup is short:
- Put the Claude app on your iPhone, signed in with the same account as Claude Code.
- In the session, type
/remote-control. - Type
/configand switch on Push when actions required. That one covers permission prompts and questions. There's also Push when Claude decides, where Claude pings you when it thinks you'd want to know.
You need a paid plan for this (Pro, Max, Ultra, Team or Enterprise). An API key won't do.
What you get on the wrist is a buzz and a line of text. There's no Claude app for the watch, so to actually answer you take out the phone. For a lot of people that's enough, and it costs nothing extra.
The do-it-yourself one: a Notification hook
Claude Code has a Notification hook that runs when it's waiting for you. Point it at a small script that posts to ntfy or Pushover, and your phone, and so your watch, gets a push.
People have published theirs. claude-watch tells you when a task finishes. claude-remote-approver goes further and puts Allow and Deny buttons on the notification, through ntfy.
This works on any plan and any computer, and you can read every line of what runs. The catch is that most of these only tell you something happened. You still walk back to the terminal.
The one I built: answer on the watch
I wanted to answer from the wrist, not just be told. So Tapproval puts the prompt itself on the watch: the command, Allow, Deny. A PermissionRequest hook on the computer hands each prompt to a small relay, and the watch talks to that.
Two things I added because I didn't trust myself to read carefully on a 40 mm screen. Every prompt gets a risk mark, one of five, worked out on your own computer. A force-push to main says CRITICAL before I've read a word. And nothing is approved for you unless you turn that on yourself; out of the box every card waits for a tap.
The part on the computer is open source, plain Python with no dependencies. There's no account, and nothing goes through a server of mine. The first 100 answers are free, then it's $5, once.
I'm not the only one doing this. There are other watch projects, some you build from source yourself. I keep a comparison page with a source for each claim, including the ones that don't flatter me.
Leave the phone at home and it changes
This is the bit I didn't think about until it bit me. The watch only shows an iPhone app's notifications while the two are connected. Walk the dog without the phone and the first two options go silent. Nothing breaks, you just don't hear anything.
A watch app can reach the computer by itself, over Wi-Fi or cellular. That's the case I actually built Tapproval for.
If you run Remote Control and a watch app together
You can, and I do. One annoyance: with Push when actions required on, the Claude app still pushes for prompts the watch already answered, so you get two buzzes for one question. It's a known bug on Anthropic's side and still open. The setup guide says which switch stops it and what you give up.
The short version
| Remote Control | Notification hook | Watch app | |
|---|---|---|---|
| Answer on the watch | No, on the phone | Only if the script adds buttons | Yes |
| Works without the iPhone | No | No | Yes |
| You need | A paid Claude plan, the Claude app | A script and a push service | The app and its helper |
| Costs | Nothing extra | Nothing, or the push service | 100 answers free, then $5 once |
| Shows how risky the command is | No | No | Yes |
If you always have your phone on you, start with Remote Control. Honestly. It's already there.
If you want to try mine
Search for "Tapproval" in the App Store on the watch or the phone (it's a watch app, there's no phone app to install). Then run the one-line install from the Claude Code guide on the computer. Open the app on the watch while both are on the same Wi-Fi and it finds the computer.
It covers Codex with the same install. I wrote about Codex's four ways last week.
Tapproval is independent. It isn't affiliated with or endorsed by Anthropic, OpenAI or Apple.
Top comments (0)