[FEATURE] Remote Control: Push notifications on iOS when permission approval is needed
Description
When using Remote Control from the iOS Claude app, there's no push notification when the CLI session needs permission approval (e.g., tool calls waiting for user consent). You have to keep the app open and actively watch the session to notice when it's blocked on a permission prompt.
Expected Behavior
The iOS Claude app should send a push notification when a Remote Control session is waiting for permission approval, so the user can step away and be alerted when their input is needed — similar to how CI/CD systems notify on pipeline approval gates.
Current Behavior
No notification is sent. The session silently blocks until you happen to check the app, which can waste significant time if you're away from your phone.
Why This Matters
Remote Control is most useful when you can kick off a task and walk away. Without notifications, you have to babysit the session to catch permission prompts, which defeats the purpose of remote control.
Environment
- Platform: iOS (Claude app)
- Feature: Remote Control
Showing cached comments. Read the full discussion on GitHub ↗
13 Comments
Every day I check and see if this is a new feature. Def a needed thing.
I wish this feature were implemented, or at least a better way to manage permissions to allow safer actions. On Android too.
It would be super useful to have this, otherwise I’m just checking my phone every few minutes
It would be fantastic!
How is this not implemented already?
Dario please, this severely limits use cases of dispatch
Tried reinstalling Claude on my iOs and also sending notification using the recent 2.1.110 PushNotification. Even resetting all settings in iPhone... No lucky 😢
It would be super useful. 10/10
This is almost critical. It is too easy to get distracted and leave Claude Code to wait for a few minutes or much more. Finally a great use of notifications.
this is the use case that makes remote control actually hands-off, agreed. being alerted is half of it, the other half is answering from the phone so the session unblocks without you returning to the desk.
a partial answer today: repowire gives every session a bidirectional telegram/slack peer with inline approve/deny buttons, so an agents question is pushed to your phone and the reply routes back. the honest gap is the integration boundary, repowire surfaces its own asks cleanly, but wiring claude codes native permission prompt into a hook so it can be relayed is the missing native piece (and related to the subagent permission-hook bypass in #23983).
so the phone channel + actionable buttons exist now, the bit only anthropic can close is exposing the permission gate to that channel. worth doing, this is one of the highest-value notification cases.
This exists — ai-agent-notifier sends push notifications to iOS (via Pushover or ntfy) when Claude Code needs approval or finishes a task:
https://github.com/user-attachments/assets/5714b528-7e04-478e-abfd-2a3d05db562c
https://github.com/DevinoSolutions/ai-agent-notifier
For anyone who needs this today while the official feature catches up — I built a third-party iOS app, PocketAI, that does exactly this: permission approvals come to your phone as notifications, and you can see the full command / a red-green diff before you approve. (Not affiliated with Anthropic.)
It drives the Claude Code on your own machine, and unlike the official Remote Control it doesn't care how you're authenticated — API key, subscription, or a gateway via ANTHROPIC_BASE_URL all work.
App Store: https://apps.apple.com/us/app/pocketai-remote-coding-agent/id6787897526
Happy to go deep on how it handles permissions.
For anyone who needs this today while the official feature catches up — I built a third-party iOS app, PocketAI, that does exactly this: permission approvals come to your phone as notifications, and you can see the full command / a red-green diff before you approve. (Not affiliated with Anthropic.)
It drives the Claude Code on your own machine, and unlike the official Remote Control it doesn't care how you're authenticated — API key, subscription, or a gateway via
ANTHROPIC_BASE_URLall work, which is the main reason I made it.Happy to go deep on how it handles the permission flow if useful.