Claude Code Remote Control connects a Claude Code session on your machine to claude.ai/code in the browser or to the Claude app on iOS and Android. You start a task at your desk and keep steering it from your phone or another computer. Everything still runs on your machine: files, tools, MCP servers and project settings stay there, while the browser and app are only a window into that session. You start Remote Control with claude remote-control, with claude --remote-control, or inside a running session with /remote-control. According to Anthropic, you need a Pro, Max, Team or Enterprise plan with a claude.ai sign-in; API keys are not supported, and on Team and Enterprise an Owner has to turn the feature on first. Your machine opens no inbound port for it. While the connection is up, however, Anthropic stores the session transcript on its servers.
I use Remote Control in my own operation: five of my AI employees run on their own Linux server, each in a tmux session with claude remote-control, and I steer them through the Claude app. That setup is described in Claude Code on a Server: tmux, Remote Control. This article explains the feature itself, for your own laptop as much as for a server. Everything about Claude is collected on my Claude page.
Remote Control or cloud session: where Claude actually runs
Remote Control and Claude Code on the web share the same interface at claude.ai/code, and that is where most of the confusion comes from. The difference is location: with Remote Control the session runs on your machine, with a cloud session it runs on infrastructure that Anthropic manages by default. Anthropic recommends Remote Control when you are in the middle of local work and want to continue from another device, and a cloud session when you want to start without local setup, work on a repository you have not cloned, or run several tasks in parallel.
The docs list several ways to work with Claude Code when you are away from your terminal; here are the most relevant for individuals and small teams, plus the cloud session:
According to the docs, cloud sessions are available on Pro, Max and Team, and for Enterprise users with certain seats; they keep running when you close your laptop and count toward your plan's usage. With Remote Control, by contrast, your machine has to stay on and the claude process has to keep running.
Requirements
According to Anthropic's docs, Remote Control needs:
- A subscription: Pro, Max, Team or Enterprise. API keys are not supported. On Team and Enterprise, Remote Control is off by default, and an Owner turns it on in the organization's Claude Code settings. With custom roles on Enterprise, you also need the permissions for cloud sessions and for Remote Control.
- A claude.ai sign-in: with
/login in Claude Code. A long-lived token from claude setup-token is not enough, according to the docs, because it can only make model requests.
- A direct route to the Anthropic API: Remote Control does not work through Amazon Bedrock, Google Cloud or Microsoft Foundry, through an enterprise Claude apps gateway, or when
ANTHROPIC_BASE_URL points to an LLM gateway or proxy.
- A trusted folder: in a folder you have not trusted yet,
claude remote-control asks first.
- Feature flags left on: if
CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC or DISABLE_GROWTHBOOK is set, Remote Control is unavailable, according to the docs.
Setting up Remote Control: three ways in the terminal
The docs describe three ways to start Remote Control from the command line:
# 1. Server mode: waits for connections, can serve several sessions
claude remote-control --name "My Project"
# 2. A normal interactive session that can also be steered remotely
claude --remote-control "My Project"
# 3. Inside a running session: the history comes along
/remote-control My Project
The first time, Claude Code asks once whether you want to enable Remote Control. In server mode, the terminal then shows a URL for the session, and pressing the spacebar shows a QR code for your phone. The difference between the modes: in server mode, the process waits for remote connections and can serve several sessions at once, up to 32 by default. With claude --remote-control you get a normal terminal session in which you can keep typing locally while it is also reachable remotely. /remote-control, or /rc for short, takes a session that is already running, history included. The same /remote-control and /rc commands also work in the VS Code extension and in the Code tab of the desktop app.
For server mode, the docs list a number of flags; these are the ones I consider most important in practice:
The spawn mode is more than a detail. With same-dir, all sessions share the folder and, according to the docs, can conflict when they edit the same files. If you work on several things in parallel in a Git repository, worktree is the safe choice.
Connecting from your phone or browser
Once the session is up, you have three ways in: open the URL shown in any browser, scan the QR code with your phone, or open claude.ai/code or the Claude app and find the session in the list. In the app it is under "Code"; Remote Control sessions show a computer icon with a green dot while they are online. If you do not have the app yet, /mobile in Claude Code shows a QR code for the app store.
From your phone you type as in a chat, attach photos or files and see subagents running in the background. You can switch the model and effort from the connected device. When Claude needs a permission or asks a question, the docs say it stays open until you answer. To hear about it, there are push notifications: in /config, turn on "Push when Claude decides" for notifications at Claude's discretion, "Push when actions required" for permission prompts and questions, or both. The phone has to use the same account and organization as the terminal.
If every interactive session should be remotely controllable automatically, set "Enable Remote Control for all sessions" to true in /config, or set remoteControlAtStartup in your ~/.claude/settings.json. One detail in the docs shows that Anthropic thought about security here: a committed .claude/settings.json in a project can turn Remote Control off, but cannot turn it on for everyone who opens the repository.
Security: what goes over the wire and who can get in
Anthropic describes the connection like this: your local session makes outbound HTTPS requests only and never opens inbound ports. It registers with the Anthropic API and polls for work. All traffic travels over TLS through the Anthropic API, using several short-lived credentials, each scoped to a single purpose. For a machine behind a firewall, that means you open nothing except outbound port 443 to the Anthropic API.
Two points are worth knowing before you use Remote Control with client data. First, while the connection is up, Anthropic stores the transcript, including your messages, Claude's responses and tool activity, to keep devices in sync and to reconnect after a drop; the data usage policy applies. Execution and file access stay on your machine. Second, access is tied to your claude.ai account: whoever takes over your account can steer your session, and it acts with your user account's privileges on that machine. What else belongs to securing Claude Code is covered in my piece on Claude Code security; the data protection questions that come up when companies use Claude are in Claude Privacy and GDPR: What Applies.
For that risk there is Trusted Devices, which the docs currently list as a beta; according to the release notes, admins on Team and Enterprise have been able to require it since 25 June 2026. According to the docs, it ties Remote Control access to a known device and a recent authentication, not just a signed-in account. With it on, a Remote Control session can only be viewed and steered from an enrolled device with a sign-in no older than 18 hours; after that you confirm with Face ID, Touch ID, Windows Hello or a passkey. On Team and Enterprise an Owner turns it on for the whole organization; on Pro and Max you turn on "Require trusted devices" yourself. According to the docs, Anthropic never receives biometric data. The disableRemoteControl key turns Remote Control off entirely; organizations put it in managed settings to block it on a device. With Zero Data Retention or the HIPAA configuration, Remote Control cannot be turned on.
My recommendation from my own operation: start remotely steered sessions in a permission mode you chose deliberately, and block dangerous tools technically rather than by asking nicely. My server sessions run with acceptEdits; anything that goes out still needs my explicit approval.
Everyday limits
- The machine has to stay on. If you close the terminal or stop the
claude process, the session goes offline. On a remote machine, the docs recommend tmux or screen so the session survives the end of the SSH connection.
- One remote session per process. Outside server mode, each Claude Code instance serves exactly one remote session.
- Network outages. In server mode, Claude Code gives up after roughly ten minutes without a network and exits. An interactive session keeps retrying and reconnects on its own.
- Bringing sessions back after a stop. After
Ctrl+C, claude remote-control, --continue or --session-id bring the sessions back for about four hours.
- Not every command works from your phone. Commands that only run in the terminal interface, such as
/plugin or /resume, work only locally. /compact, /clear, /context or /model sonnet work remotely too.
Frequently asked questions
What is Claude Code Remote Control?
A feature that lets you steer a Claude Code session on your machine from the browser at claude.ai/code or from the Claude app. The session keeps running on your machine; the browser and app only display it and send your messages.
How do I set up Claude Code Remote Control?
Sign in through claude.ai with /login, run claude remote-control in your project folder and confirm the one-time question. Then open the URL shown or scan the QR code. In a running session, /remote-control is enough.
Which plans do I need, and does Remote Control cost extra?
Pro, Max, Team or Enterprise with a claude.ai sign-in; API keys are not supported. Anthropic names no separate price. On Team and Enterprise, an Owner has to turn Remote Control on first. If ANTHROPIC_API_KEY is set, remove the variable and sign in through claude.ai.
What is the difference from Claude Code on the web?
With Claude Code on the web, the session runs in the cloud, Anthropic-managed by default, and keeps working when your laptop is closed. With Remote Control it runs on your machine, with your local files, tools and MCP servers, and your machine has to stay on.
Is Remote Control secure?
Your machine opens no inbound port, and traffic travels encrypted through the Anthropic API. The transcript is stored at Anthropic while you are connected, and without further safeguards, access depends on your signed-in account. Trusted Devices ties access to enrolled devices and a recent sign-in.
Where to go from here
Try it on your own machine first: run /remote-control in a running session, connect with your phone, and turn on notifications for permission prompts in /config. If that works for you and you want sessions to run permanently, the next step is a server with tmux, as described in Claude Code on a Server: tmux, Remote Control. The entry point for Claude Code itself is Claude Code: A Guide for Beginners. How others lead their AI employees on the go is what we discuss in my German-speaking community Claude Practitioners.
