Approve Claude Code permissions from your phone

You start a long task, walk away, and Claude stops at "Do you want to proceed?" on a computer you're not at. Here's how to answer from your phone, and how to make Claude ask less without handing it the keys to everything.

The short answer

1. Run the session with Remote Control and open it in the Claude app under Code. Permission prompts and Claude's questions show up there, and you tap to answer.
2. Run /config and turn on Push when actions required, so your phone tells you when Claude is waiting.
3. Cut the number of prompts with allow rules in your settings and a starting mode such as Accept edits or Auto. Add deny or ask rules for anything you never want run unattended.
4. If a prompt doesn't show up on the phone, it's still waiting on the computer. That's a known bug (below), not your setup.

What people report

• #29214: "mobile app shows permission prompts despite --dangerously-skip-permissions". Replies run into October 2026. One person says claude remote-control --permission-mode bypassPermissions fixed it; another tested the same command and says it didn't.
• #35637: "permission prompts not rendering on mobile app". The phone shows nothing, the session looks frozen, and the terminal on the computer is waiting for an answer. Later replies reproduce it on v2.1.299, and between two desktop apps.
• #28508: a multiple-choice question answered on the phone never reaches the CLI, so the session waits forever.

Answering from the phone works for many people, but not every time. Plan for that.

Step 1: get the prompts onto your phone

Remote Control needs a claude.ai login on the computer (Pro, Max, Team or Enterprise; API keys aren't supported). Start a session with it:

claude --remote-control

Or type /remote-control in a session that's already running. Then open the Claude app, go to Code, and pick the session. Per the Remote Control docs, Claude Code keeps permission prompts and questions open until you answer them. Other dialogs it forwards close after five minutes by default (the dialogExpiry setting).

Now make the phone tell you. In the terminal, run /config and turn on Push when actions required (permission prompts and questions). Allow notifications for the Claude app when your phone asks.

If /config shows No mobile registered, open the Claude app on the phone once. On iOS, Focus modes can hold pushes back; on Android, exempt the app from battery optimization.

Step 2: change the mode from the phone

In the app, tap + in the prompt box, then Permission. A Remote Control session offers Manual, Accept edits, Plan and Auto. You can't pick Bypass permissions from the app. Changes on either side stay in sync.

To pick the starting mode on the computer instead:

claude remote-control --permission-mode acceptEdits

What runs without asking, per the permission modes docs:

• Manual (config value default): reads only.
• Accept edits: reads, file edits, and common filesystem commands like mkdir, mv and cp.
• Plan: reads. Edits wait until you approve a plan.
• Auto: everything, with a separate classifier model reviewing each action before it runs. The docs are clear that it "reduces permission prompts but does not guarantee safety".

Step 3: pre-approve the boring commands

Most phone taps are for the same few commands. Put them in ~/.claude/settings.json (every project on that computer) or in .claude/settings.local.json inside one repository (only you, kept out of git):

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

How the rules behave, from the permissions docs:

• Deny is checked first, then ask, then allow. A narrower allow rule can't carve an exception out of a deny rule.
• Put the * after the subcommand. Bash(git log *) allows only git log; Bash(git *) allows every git command, push included.
• An ask rule prompts in every mode, Auto included. That's how you make sure a push or a deploy always comes to your phone for a yes.
• Tapping "Yes, and don't ask again" on a Bash prompt saves the rule to .claude/settings.local.json at the repository root, so your answers add up over time.
• /permissions lists every rule and the file it came from.

Safe defaults for unattended work

• Start in Accept edits or Auto, not bypass. The docs say to use bypass only "in isolated environments like containers, VMs, or dev containers without internet access", and on Linux and macOS it refuses to run as root.
• Keep git as your undo button. Work on a branch and let Claude commit often, so a bad run is one git reset away.
• Expect a few prompts anyway. Ask rules, Claude's questions, and deletes of critical paths such as rm -rf ~ still come to you in every mode. That's by design.

When a prompt never shows up on the phone

If the app shows Claude "working" for a long time with nothing to tap:

1. Check the computer. If the terminal is sitting at a prompt, you've hit #35637 or #28508. Answer it there, or press Esc.
2. From the app, switch the session to Accept edits or Auto, so fewer actions need a prompt in the first place.
3. For a backup signal, use a hook. A PermissionRequest hook runs on the computer the moment Claude is about to ask, whether or not the app draws the prompt, so it can send you an alert through a tool you already use.

Related guides

• A tmux alternative for keeping Claude Code running: where the session lives while you're away.
• Tailscale + tmux + Termius, step by step: the do-it-yourself phone setup.
• Claude Code Remote Control keeps disconnecting: when the app says "disconnected".
• Running Claude Code on a server: moving Claude off the laptop.

If you'd rather not keep a computer on for this

Prompts only reach your phone while the machine running Claude is on and connected. Podway is an always-on cloud computer for Claude Code. You sign in with your own Pro or Max plan and open the session from the Claude app via Remote Control. Each pod is its own isolated machine, so Claude runs there without permission prompts: there is nothing to approve from your phone, and you watch and steer the work from the Claude app instead. A pod never sleeps, and if the agent crashes, Podway restarts it and Remote Control reconnects. Before your Claude login expires, Podway tells you, with a one-click reconnect. On Team or Enterprise, your org Owner has to enable Remote Control first.

Your first pod is free, no card needed, and pods start at $4/month. Start a pod, or read more on the agent computer page.