Approve Claude Code permissions from your iPhone

Updated

Quick answer

You can approve a Claude Code permission request from your iPhone because Claude Code runs a PermissionRequest hook when a tool needs a permission, and that hook can return the decision. Monish uses it: the request reaches your lock screen with the exact command, you tap Allow or Deny, and Claude carries on. The prompt stays on screen in your terminal too, and the first answer wins. Your deny rules apply whatever you tap.

Why a permission prompt stalls a task

In Manual mode, Claude Code stops and asks before most commands, file edits and network requests. A permission prompt never resolves on its own: if you are not there, the whole task waits, sometimes for hours. Since version 2.1.283, interactive sessions start in auto mode, where a classifier reviews actions instead of you and prompts become rare. Many people still work in Manual mode on code they care about, and even auto mode keeps a prompt for anything your ask rules name.

Anthropic offers an official way out: with a Claude plan and a session connected to Remote Control, Push when actions required in /config sends permission prompts and questions to the Claude app. If that works for you, it is enough. The rest of this page is the other road, which needs no plan and no sign-in.

How a hook hands back your decision

PermissionRequest runs just before Claude Code would show a permission prompt, and only then, unlike PreToolUse, which runs before every tool call. It receives the tool and its input as JSON:

What the hook reads on standard input (shortened)

{
  "hook_event_name": "PermissionRequest",
  "permission_mode": "default",
  "cwd": "/Users/you/dev/harbor",
  "tool_name": "Bash",
  "tool_input": { "command": "npm install && npm run build" },
  "permission_suggestions": [ ... ]
}

It answers by printing one JSON object and exiting 0:

Allow, or deny with a reason Claude reads

{"hookSpecificOutput":{"hookEventName":"PermissionRequest","decision":{"behavior":"allow"}}}
{"hookSpecificOutput":{"hookEventName":"PermissionRequest","decision":{"behavior":"deny","message":"Denied from my phone."}}}

If the hook prints nothing, the usual prompt decides. A hook that wants to ask someone elsewhere simply waits before printing: Claude Code gives a command hook 600 seconds by default. Two rules keep this safe. An allow never overrides a deny or ask rule from your settings. And exit code 2, which blocks other events, is ignored here: a deny has to come through the decision object. The PermissionRequest hook has the full reference and a minimal script you can try without any app.

Set it up

  1. Install Monish

    Free on the App Store, for iOS 17.2 or later, with no account. The app shows a 6-digit code.

  2. Pair the computer

    In a terminal, run curl -fsSL https://monish.cc/install.sh | sh -s with the code. The command adds the hooks to ~/.claude/settings.json and saves a backup first. The setup guide shows each step.

  3. Work in a mode that asks

    Press Shift+Tab until the status bar shows ⏸ manual mode on, or start with claude --permission-mode manual. In auto mode you will rarely see a request.

What you see on the lock screen

The title says what Claude wants to do: Run this command? for a shell command, Edit this file?, Read this file?, Fetch this page?, or Use this tool? for an MCP tool. The body is the command or path in full, and the line under it names the project, so you know which session asks. When Claude wants to leave plan mode, the request reads Go ahead with this plan?: tap it to read the whole plan in the app, then allow it, or deny it and say what to change. Press and hold the notification for two more choices: Allow for this session, which lets the same kind of request through for the rest of the session, and Always allow this command, which lets that exact command run without asking again, in that project only. You can take it back in the app, on the computer’s page under Setup › Computers. In the app, Deny with a reason… sends Claude a line on what to do instead. A plan gets none of these: you approve it, or say what to change. If the session’s card is on your lock screen, it turns to the same request, with Allow and Deny on the card, and a notification with sound goes out. Once you tap, the card shows Allowed from iPhone or Denied from iPhone with the time.

If you don’t answer, or you are at your desk

Nothing is decided for you. Your iPhone can answer within the answer window, a single setting in the app’s Settings: 10 minutes by default, or 2, 5 or 30 minutes, or 1 hour. After that, nothing is allowed or denied: Claude Code asks you in the terminal, and the request shows as Expired on your iPhone. In a non-interactive claude -p run there is no prompt to fall back on, so a request nobody answers is denied.

At your desk, nothing changes: while you’re typing in the terminal, Claude Code asks there as usual, and the request reaches your iPhone as a silent note, with the same buttons, in case you step away. Since the prompt stays on screen there the whole time, you can also answer it at the keyboard after the push went out. The first answer wins, and the card on your iPhone closes as Answered in the terminal.

Approve less often: allow rules

The best remote approval is the one you never need. Commands you run every day belong in allow rules, and the risky ones in ask or deny rules, in ~/.claude/settings.json or through /permissions:

Fewer prompts, same control

{
  "permissions": {
    "allow": ["Bash(npm run test *)", "Bash(npm run lint)", "Bash(git commit *)"],
    "ask": ["Bash(git push *)"],
    "deny": ["Read(./.env)"]
  }
}

Choosing Yes, and don’t ask again at a prompt saves the same kind of rule. Rules are checked deny first, then ask, then allow. Permission modes, and fewer prompts goes further.

Questions

Can a tap on my iPhone allow something my settings deny?
No. Claude Code evaluates deny and ask rules whatever a hook returns, so an Allow from the phone never overrides a deny rule.
What happens if I answer in the terminal first?
Claude Code applies your terminal answer straight away. The first answer wins, and the card on your iPhone closes as Answered in the terminal.
Does it work in auto mode?
Auto mode asks for few permissions, so there is little to approve. Claude’s questions and the end of each turn still reach your iPhone.
Do I need Remote Control or a Claude plan?
No. The hook runs on your computer, so it works with a Claude plan, an API key or a cloud provider, with no claude.ai sign-in.

More guides

New to Monish? The home page shows what lands on your iPhone, and all guides are listed here.

Get Monish on your iPhone

Free, no account. Your 6-digit code is waiting in the app.

Download on the App Store