Claude Code Remote Control: run it from your phone and watch

Remote Control is Anthropic's own way of leaving the desk while Claude Code keeps working. You start a session on your machine, scan a QR code, and the session appears in the Claude app on your phone or at claude.ai/code, where you can read what it is doing, send it messages and approve its permission requests. This guide covers what it is, how to set it up, what it needs, what it does not do, how the security works, and what a watch adds on top. Everything below is checked against Anthropic's documentation as of October 2026.

What is Claude Code Remote Control?

Remote Control connects claude.ai/code, or the Claude app for iOS and Android, to a Claude Code session running on your computer. The documentation's one-line summary: "Start a task at your desk, then pick it up from your phone on the couch or a browser on another computer."

The important word is running. The session stays on your machine the whole time, with your filesystem, your MCP servers, your tools and your project configuration. The phone and the browser are "a window into that local session". That is the difference from Claude Code on the web, where the session runs on cloud infrastructure: Remote Control is your laptop, driven from elsewhere.

The claude remote-control subcommand first appears in the Claude Code changelog at version 2.1.51, in 2026, with a wider rollout a few releases later.

What plans and setup does it need?

• A claude.ai subscription: Pro, Max, Team or Enterprise. API keys are not supported; if ANTHROPIC_API_KEY is set, Remote Control refuses with a message naming it.
• On Team and Enterprise, an Owner must first switch on the Remote Control toggle in the organisation's Claude Code admin settings. It is off by default there.
• A direct connection to Anthropic's API. It is not available through Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry, nor when ANTHROPIC_BASE_URL points at a gateway or proxy.
• A signed-in CLI: run claude and /login with your claude.ai account.
• A trusted directory. The first run in a new folder asks you to trust it.

Organisations with Zero Data Retention or the HIPAA configuration cannot enable it, because the session transcript is stored on Anthropic's servers while it is connected (more on that below).

How to set it up step by step

1. Update Claude Code: claude update.
2. Sign in: run claude, then /login, and choose the claude.ai option.
3. Install the Claude app on your phone and sign in with the same account. Inside Claude Code, /mobile shows a QR code that opens the right app store.
4. In your project directory, run claude remote-control. The first time, it explains what it does and asks Enable Remote Control? (y/n). Answer y.
5. The terminal prints a session URL. Press the spacebar to show a QR code; scan it with your phone to open the session in the Claude app, or open the URL in any browser.
6. Optionally, turn on push: run /config in Claude Code and enable Push when actions required, so permission prompts and questions arrive as notifications.
7. Send a message from the phone. You should see the tool activity in the terminal and the reply on the phone.

If you are already mid-session and want to carry on from the phone, type /remote-control (or /rc) inside the session instead; it carries your conversation history over. To stop, press Ctrl+C on the server; sessions it was serving can be brought back with claude remote-control in the same directory for about four hours.

Three ways to start it

Server mode takes useful flags: --name "My Project" sets the title you see in the app; --spawn worktree gives each on-demand session its own git worktree so parallel sessions cannot collide; --permission-mode acceptEdits sets the starting permission mode. The Desktop app and the VS Code extension use /remote-control too.

What it does, and what it does not do

It does:

• Show the conversation live, and keep it in sync, so you can send messages from the terminal, the browser and the phone interchangeably.
• Forward permission prompts and questions, and keep them open until you answer.
• Take photos and files from the phone; Claude sees attached photos directly.
• Show a diff of the changes in the session's git repository.
• Let you change the model and effort level from the device.
• Reconnect on its own if your laptop sleeps or the network drops.

It does not:

• Run when your computer is off or the claude process has stopped. The documentation is explicit: "your computer has to stay on and the claude process has to keep running." On a remote box, start it inside tmux or screen.
• Support API-key authentication, Bedrock, Vertex or Foundry, or a proxied base URL.
• Serve more than one remote session per interactive process; use server mode for several.
• Run terminal-only commands from the phone, such as /plugin or /resume.
• Give you a watch. There is no Wear OS or watchOS surface; the ask reaches your pocket, not your wrist.

How does the security model work?

The session makes outbound HTTPS requests only and never opens an inbound port on your machine. It registers with Anthropic's API, polls for work, and when you connect from another device the server routes messages between that device and your local session over a streaming connection. All of it goes over TLS, using "multiple short-lived credentials, each scoped to a single purpose and expiring independently".

Two things are worth knowing before you enable it. First, while connected, the session transcript, including your messages, Claude's replies and tool activity, is stored on Anthropic's servers to keep devices in sync and allow reconnection; execution and filesystem access stay local, and stored transcripts fall under the data-usage policy. Second, Remote Control does not change your permission rules: a deny rule holds, an ask rule prompts, and the prompt is simply forwarded to the phone. The gate is the same gate, which is the point.

For a stricter setup, Trusted Devices (in beta) ties access to an enrolled device and a sign-in no more than 18 hours old, refreshed with Face ID, Touch ID, Windows Hello or a passkey. Biometric checks run on the device; Anthropic stores only the device's public key and basic metadata. Administrators can switch Remote Control off entirely with the disableRemoteControl setting.

Remote Control versus cloud sessions

Anthropic also lists Dispatch (message a task from the app to the Desktop app), Channels (Telegram, Discord or your own server pushing events into a session) and scheduled tasks as ways to work away from the terminal. Remote Control is the one for a session that already exists.

What a watch adds on top

Remote Control solves the first half of the problem: the ask reaches you. It still arrives on a phone, which means a pocket, an unlock, an app and a screen you then read, and a phone in hand pulls you into everything else on it. A watch changes three things.

• One-tap approve, with the arguments in view. PeakWatch shows the tool, its arguments and the SDK's reason on the watch face; Allow, Deny and Say why are one tap, and Always covers the repeats for the rest of the session.
• Auto-deny. A sixty-second ring drains on the bezel and the relay denies on your behalf when it ends, so an ask you miss is never a yes. Remote Control keeps forwarded permission prompts open until you answer them.
• No phone in hand. The watch talks to a relay on your machine over WiFi or LTE on its own, so the phone stays in the bag and the laptop stays shut.

PeakWatch is a separate layer, not a Remote Control client: a small relay you run drives Claude Code through the Claude Agent SDK, and it mirrors sessions you started in a terminal too. It is in private beta on Apple Watch and Wear OS.

Use both: Remote Control for the full conversation and the diff on a phone screen, the watch for the decision that arrives while your hands are busy. The habits that make long sessions safe to leave and the case for supervising asynchronously are the other two pieces.

Hen Eytan · Founder, UPeak

Builds the UPeak apps and writes up the research behind them.