Claude Code Remote Control: Run Sessions From Your Phone

Claude Code Remote Control connects the Claude mobile app or claude.ai/code to a Claude Code session running on your own computer, so you can steer local work from your phone. Run claude remote-control in your project folder (or /remote-control inside a session), scan the QR code, and the session stays on your machine with your files and tools while the phone acts as the screen and keyboard.

Checked October 2026 against Anthropic's Remote Control docs, Claude Code on mobile, cloud sessions docs and the CLI reference. Flags and menus change between versions; the docs are the reference if they differ from this page.

By the end of this tutorial you will have a Remote Control session you can reach from your phone, push notifications for approvals, a safe default permission setup, and a handoff routine for the day you are away from your desk. Everything here follows the official docs; there are no tunnels, no port forwarding and no API keys on your phone, because the feature needs none of them. If you are new to the agent itself, read our Claude Code complete guide first.

Before you start: what Remote Control needs

Remote Control is available on Pro, Max, Team and Enterprise plans. API keys are not supported, so a Claude Console login will not work. On Team and Enterprise, an Owner has to switch on the Remote Control toggle in Claude Code admin settings first.

If you do not have the app yet, run /mobile inside Claude Code: it shows a QR code for claude.ai/mobile, which opens the right app store for your phone.

How to set up Claude Code Remote Control

You can start a Remote Control session from the CLI, the VS Code extension or the Code tab of the desktop app. The terminal route gives you the most control, so the steps below use it.

The other ways to start are useful in different moments. Mid-task, /remote-control (or /rc) carries over your current conversation history; claude --remote-control "My Project" starts a normal interactive session you can type into locally and remotely. To connect every interactive session automatically, set Enable Remote Control for all sessions in /config; the docs note that auto-connect signs in with your own account, so the session appears only in your own Claude apps.

Using Claude Code from your phone

Once connected, the phone shows the conversation as it happens, including subagents and workflows already running in the background; stopping one from the phone stops it on your machine. You can attach photos and files from the phone: Claude sees photos directly, and Claude Code downloads other files to your machine as @ references. Typing @ autocompletes file paths from your local project. If the session's folder is a Git repository, a diff pane shows your changes.

Many commands work from the phone, including /compact, /clear, /context, /usage and /model or /effort with an argument, such as /model sonnet. Commands that only run in the terminal interface, such as /plugin and /resume, stay local. For approvals, permission prompts and questions stay open until you answer them, which is why the push setting for actions required matters most.

How Remote Control keeps your machine safe, and what never to expose

The design is the security story. Per the connection and security section of the docs, your local session makes outbound HTTPS requests only and never opens inbound ports. It registers with the Anthropic API and polls for work; your phone talks to the Anthropic API, which routes messages to your session over a streaming connection. All traffic uses TLS, and the connection uses multiple short-lived credentials, each scoped to one purpose.

Beyond that, the risk to manage is what the agent can do while you are not watching the terminal. These are the controls the docs give you:

• Pick a conservative permission mode. Server mode accepts --permission-mode, for example acceptEdits, which auto-accepts file edits but still asks before other terminal commands. Avoid bypass mode for sessions you steer from a phone, and answer prompts as they arrive.
• Sandbox the sessions. The docs say you can sandbox the sessions claude remote-control starts by turning on sandboxing in a settings file.
• Turn on Trusted Devices. In beta on all four paid plans, it requires an enrolled device and a sign-in no older than 18 hours before anyone can view or steer a Remote Control session. On Pro and Max you switch on Require trusted devices yourself; biometric checks stay on the device.
• Revoke lost devices. Remove a lost or stolen phone from the Trusted devices section of your claude.ai account settings; that revokes its credential immediately.
• Know what is stored. While connected, the session transcript, including messages, responses and tool activity, is stored on Anthropic servers under the data usage policy. Do not paste secrets into the chat from your phone.
• Turn it off where it does not belong. The disableRemoteControl setting switches the feature off entirely, and a project settings file with false blocks auto-connect for that repository; a checked-in true is ignored.

The same rule applies to cloud sessions. Environment variables on a cloud environment can be read by anyone who uses that environment, so the docs recommend network secrets (on Pro and Max) for API keys, which the agent proxy adds to requests without the session seeing the key.

Session handoff patterns

Remote Control is about moving one piece of work between screens without losing context. These are the handoffs the docs support:

Two details are worth knowing. Server mode runs many sessions from one process, and --spawn worktree gives each on-demand session its own Git worktree, so parallel tasks do not edit the same files; you can press w to toggle it while the server runs. And the docs say handoff from the CLI to the cloud is one-way: you can pull cloud sessions into your terminal with --teleport, but you cannot push an existing terminal session to the cloud. If you would rather manage sessions in a window than in terminals, the Claude Code desktop app runs /remote-control too.

A docs-based workflow for a day away from your desk

Here is a generic routine assembled from what the docs describe, for a day when you will be on the move but want work to keep going:

1. Before you leave: commit your work, then decide where each task should run. Anything that needs your local database, tools or MCP servers stays local under Remote Control. Anything that only needs the GitHub repo goes to a cloud session, which keeps running with the laptop closed.
2. Start the local side: keep the computer awake and plugged in, then run claude remote-control --spawn worktree --permission-mode acceptEdits. On a remote Linux box reached over SSH, start it inside tmux or screen so it survives your SSH session ending, as the docs suggest.
3. Queue clear tasks: from the phone, give each session one specific job with a test that proves it worked. Ask for a push when it finishes, for example "notify me when the tests finish".
4. On the move: answer permission prompts from the notifications. Use /usage and /context from the phone to keep an eye on limits and context.
5. Back at the desk: review the diffs properly on the big screen before you merge anything. Phone screens are fine for steering, poor for code review.

Claude Code web: Remote Control or a cloud session?

Both live at claude.ai/code and in the Code tab of the Claude app, which is why people mix them up. The difference is where the code runs. Cloud sessions, also called Claude Code on the web, run on cloud infrastructure that Anthropic manages by default and need your repository on GitHub; they are available on Pro, Max and Team, and on Enterprise premium or Chat + Claude Code seats. Remote Control runs on your machine.

Dispatch is a third option on Pro and Max: you message a task from the phone and the desktop app on your computer decides whether to run it as a Code session. Our VS Code setup guide covers the editor side, which also supports /remote-control.

Troubleshooting Remote Control

• "Remote Control requires a claude.ai subscription". You are signed out, or an API key or token is taking precedence. Run claude auth login, choose the claude.ai option, and remove ANTHROPIC_API_KEY or ANTHROPIC_AUTH_TOKEN if the message names one.
• "Only available when using Claude via api.anthropic.com". ANTHROPIC_BASE_URL points at a gateway or proxy, or you use a cloud provider. Unset the variable for this session.
• "Isn't enabled for this account". Usually cached entitlements after a plan change. Run claude auth logout, then claude auth login, and update Claude Code. claude doctor shows which eligibility check failed; an organization policy block has its own message, which on Team or Enterprise means asking an Owner.
• "Workspace not trusted". Run claude remote-control in an interactive terminal once and answer the trust question.
• Session offline. The claude process stopped or the computer is off. In server mode, a long network outage ends the process after roughly 10 minutes; run it again.
• No push notifications. If /config shows No mobile registered, open the Claude app so it refreshes its push token. Check Focus modes on iOS and battery optimization on Android.

Remote Control is one piece of running an AI-assisted build without being chained to a desk. If you want the full path from idea to a shipped app with Claude Code and similar tools, our AI SaaS Builder program walks through it step by step. For first-time setup on your computer, start with the Claude Code install guide.

Claude Code Remote Control: FAQ