Approving Claude Code away from your desk: the options compared
Written by the maker of Tapproval, so read it with that in mind. Every claim about another product links to its source; facts as of 25 September 2026.
Claude Code stops and waits whenever it needs permission. Several tools now let you answer from somewhere other than the terminal. They differ more than their descriptions suggest: some need an account and a server in the middle, some need a particular plan, only a few reach the Apple Watch, and some also drive Codex.
Side by side
| Tapproval | Claude Remote Control | Happy | Omnara | claude-watch | |
|---|---|---|---|---|---|
| Price | 100 answers free, then $5 once | Free app; needs a Pro, Max, Team or Enterprise plan | Free; long voice sessions paid | No platform fee; managed compute paid | Free |
| Apple Watch | Watch app, built for the wrist | No watch app; the iPhone notification is mirrored | No | Yes, in the iOS app | Watch and iPhone app |
| iPhone / Android / web | Watch only | iOS, Android, web | iOS, Android, web, Mac | iOS; others not checked | iPhone |
| Account | None | claude.ai account | Created automatically, no email | Required | None |
| Server in the middle | None of ours: direct, or an encrypted tunnel | Anthropic's | Hosted relay, self-hostable | Hosted, self-hostable | None (local network only) |
| Open source | The computer half | No | Yes (MIT) | Yes (Apache-2.0) | Yes (MIT) |
| Codex | Not yet | No | Yes | Yes | No (a fork targets Codex) |
| Install | App Store + one sentence to Claude Code | Built in | App store + happy command | App store + login | Build it yourself in Xcode |
| Risk shown on each prompt | Yes, five tiers | No | No | Not checked | No |
Which one fits
- You already pay for Claude Pro or Max and mostly carry your phone: start with Anthropic's Remote Control. It is built in and needs nothing installed.
- You also use Codex, or want Android or a desktop client: Happy or Omnara. Happy is the most widely used. Omnara has moved towards managed agents in its own cloud.
- You want to build your own: claude-watch is a good open-source start, and watch-control covers Codex over Tailscale.
- You want the watch to be the whole interface, with no account, no server of ours in the middle, and each prompt marked with how risky it is: that is what Tapproval is for. By default it approves nothing on its own; every card waits for your tap.
How the approval actually travels
All of these rely on the same thing underneath: Claude Code's hooks, or its own Remote Control channel. A tool that answers from outside the terminal uses the PermissionRequest hook, and the decision it prints has a shape that is easy to get wrong. A wrong shape is ignored without an error. How the hook works covers the exact shape.
Sources
- Remote Control requirements and platforms: code.claude.com/docs/en/remote-control; notifications mirrored to the watch: claude-code#96126
- Happy: github.com/slopus/happy, self-hosting, paid voice
- Omnara: github.com/omnara-ai/omnara, pricing, App Store
- claude-watch: github.com/shobhit99/claude-watch; watch-control: github.com/CryptoPilot16/watch-control
Something here out of date or unfair? Write to tapproval@thoughtfulsteward.org and it will be corrected.