Claude Code Remote Control: how it works and how to keep it online

What Remote Control is

Remote Control connects claude.ai/code, or the Claude app on iOS and Android, to a Claude Code session running on your own machine. Claude keeps running locally the whole time. Code execution and filesystem access stay there, along with your MCP servers, tools and project configuration. The web and mobile screens are a window into that local session, so your computer has to stay on and the claude process has to keep running (Claude Code docs).

That one rule explains almost every Remote Control problem you'll run into. The facts below come from Anthropic's own Claude Code docs and changelog. Where we lean on user reports (the two offline messages Anthropic doesn't publish) or give our own suggestion, we say so.

Once it's connected, the other device gets your full local environment. Files, MCP servers, tools and project settings are all there. The conversation stays in sync across your terminal, browser and phone, and you can send messages from any of them. Working in a git repository? A connected device also shows a diff of your changes. From your phone you can attach photos and files, either in the Claude app or at claude.ai/code. And if your laptop sleeps or your network drops, Claude Code reconnects automatically when your machine comes back online.

The Claude Code command line, the Claude Desktop app or the VS Code extension can host a session. Anthropic's current documentation no longer labels the feature a preview.

Remote Control, cloud sessions, Channels, Dispatch or SSH: which one do you mean?

"Claude Code remote" can mean five different things, and Anthropic's docs separate them by where Claude runs:

• Remote Control (this page) drives a session that runs on your machine, from claude.ai/code or the Claude app. The machine must stay on.
• Cloud sessions (Claude Code on the web) run on Anthropic-managed infrastructure by default, so they keep going after you close your laptop. They typically work from a GitHub repository and stop after a period of inactivity (Claude Code on the web).
• Channels push events from Telegram, Discord or your own server into a session running on your machine. It's a research preview, and events only arrive while the session is open. Our guide: control Claude Code from Telegram.
• Dispatch lets you message a task from the Claude mobile app to the Claude Desktop app on your computer, and Dispatch decides how to run it. It needs a Pro or Max plan and, like Remote Control, it drives your own machine, so that machine has to stay on (Claude Code on mobile). Remote Control is for steering a session you already have running.
• SSH and tmux means logging into a machine that stays on and running Claude Code inside tmux. No Anthropic remote-access feature is involved. See how to keep Claude Code running 24/7.

What you need before you start

The docs list four requirements. You need a Pro, Max, Team or Enterprise plan, and on Team and Enterprise an Owner has to switch on the Remote Control toggle in Claude Code admin settings first. It's off by default there. You also need to be signed in through claude.ai: run claude, then /login. API keys are not supported.

Your traffic has to go straight to Anthropic. Remote Control doesn't work on Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry. Two more setups block it: ANTHROPIC_BASE_URL pointing at a host other than api.anthropic.com (an LLM gateway or proxy), and an enterprise Claude apps gateway sign-in. And your organization has to allow it. Organizations with Zero Data Retention can't enable it, and the docs say a HIPAA configuration blocks it too.

Two things catch machines that were set up for scripts or CI. Setting ANTHROPIC_API_KEY, an apiKeyHelper or ANTHROPIC_AUTH_TOKEN turns Remote Control off even when a claude.ai login also exists (Claude Code changelog, 2.1.139, 11 May 2026). And a token from claude setup-token, or CLAUDE_CODE_OAUTH_TOKEN, can only make model requests, so it can't establish sessions. The messages you'll see, and what fixes them:

• Remote Control requires a claude.ai subscription. You're signed out, and this is the wording from /remote-control or --remote-control. Run claude auth login and choose claude.ai.
• You must be logged in to use Remote Control. Remote Control is only available with claude.ai subscriptions. The same problem, in the wording claude remote-control uses. Run claude auth login.
• Remote Control requires claude.ai subscription auth. An API key, token or apiKeyHelper is in use, and the message names it. Remove it where it's set.
• Remote Control requires a full-scope login token A claude setup-token token or CLAUDE_CODE_OAUTH_TOKEN is in use. Run claude auth login.
• Remote Control isn't enabled for this account The check for your account came back off, usually because cached account data is out of date after a plan change. Run claude auth logout, then claude auth login, and update Claude Code. claude doctor shows which eligibility check failed.
• Remote Control is disabled by your organization's policy Check these in order. First, look for a disableRemoteControl setting in managed settings, which the error names. Second, see whether you're on a Pro or Max plan but still signed in under an earlier Team or Enterprise organization (run /status, then claude auth logout and claude auth login). Third, check for a HIPAA configuration: the message doesn't tell you to contact your admin, the Owner's toggle is greyed out, and the docs say to contact Anthropic support. Last, look for a Team or Enterprise Owner who has not switched the Remote Control toggle on.
• Remote Control is only available when using Claude via api.anthropic.com Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry or an ANTHROPIC_BASE_URL that points at another host is routing the session. The message names what did it. Unset that variable wherever it's set, and restart the session.
• Remote Control requires feature-flag evaluation A variable that turns off feature-flag evaluation is set, and the message names it. For CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC or DISABLE_GROWTHBOOK, unset it wherever it's set.
• Remote credentials fetch failed You're not signed in, or a firewall or proxy is blocking outbound HTTPS on port 443. If Session creation failed also appears, the subscription may be inactive. claude remote-control --verbose shows the full error.

How to start a Remote Control session

There are three ways to start it from the command line, and the Claude Desktop app and the VS Code extension have their own. Run these in your project directory:

Inside a session that's already running, this carries over the conversation so far:

/rc is short for /remote-control, --rc for --remote-control, and claude rc for claude remote-control. In the VS Code extension, type /remote-control or /rc in the prompt input. In the Claude Desktop app, type it in a local session.

Server mode is our pick for a machine you leave running, because one process serves several sessions (32 at once by default). Press the spacebar and it shows a QR code to scan with your phone. The first time, it asks Enable Remote Control? (y/n) and, in a directory you haven't trusted yet, Trust <directory>? [y/N]. Answer y to both. Trust gets saved, except in your home directory: there it's never saved, so the question comes back on every run.

To connect from another device, open the session URL in any browser or scan the QR code. Or open claude.ai/code or the Claude app and pick the session by name. In the mobile app, tap Code in the navigation. An online session shows a computer icon with a green status dot. No app yet? Run /mobile inside Claude Code and it shows a QR code to the download page.

These server-mode flags are the ones you'll actually use:

• --name "My Project" sets the session title in the claude.ai/code list.
• -c, --continue brings back the session the last server in this directory started with (v2.1.200 or later).
• --session-id <id> brings back one session by its ID (v2.1.200 or later).
• --spawn <mode> chooses how sessions are created: same-dir (the default), worktree (needs a git repository) or session (one session only).
• --capacity <N> caps how many sessions run at once. The default is 32.
• --permission-mode <mode> sets the starting permission mode for the server's sessions.
• --verbose shows detailed connection and session logs.

To turn it on for every interactive session instead, run /config and set Enable Remote Control for all sessions, or set remoteControlAtStartup in ~/.claude/settings.json.

What happens when the computer sleeps, the terminal closes or the network drops

It depends on how it stopped. Here's each case, with what the docs and changelog say:

• The laptop sleeps or the network drops. Claude Code reconnects automatically when the machine comes back online. The Remote Control docs don't say whether work carries on while the machine sleeps. Anthropic's Projects docs say a thread running on your computer pauses while that computer is asleep, so plan on nothing progressing until it wakes. The 24/7 guide covers sleep and locked screens in detail.
• You close the terminal, quit the Desktop app or VS Code, or stop the claude process. The session goes offline until you bring it back. Since Claude Code 2.1.236 (19 August 2026) it is marked offline within seconds. A session you started with claude --remote-control or /remote-control comes back with claude --continue or claude --resume.
• The machine is awake but has no network, in server mode. Claude Code gives up after roughly 10 minutes and the claude remote-control process exits. Run claude remote-control again to start a new session.
• The same, in an interactive session. You keep working locally and Claude Code retries. The docs say it retries for as long as the outage lasts. The changelog (2.1.232, 13 August 2026) says it keeps reconnecting for about 30 minutes. If it disconnects with could not reach the Remote Control server for about 30 minutes, run /remote-control.
• You stop server mode with Ctrl+C. Its sessions stop responding from your phone or browser. They aren't archived, unless another claude remote-control ran in the same directory or you started with --no-create-session-in-dir.
• A session in server mode crashes. Send it a message from a connected device and Claude Code serves it again, with no server restart (v2.1.238 or later).

After a stop, these work for about four hours, in the same directory: claude remote-control brings back every session, claude remote-control --continue brings back the session the server started with, and claude remote-control --session-id <id> brings back one. After four hours, start a new session.

Two more limits are easy to hit on a machine nobody's watching.

The first is staying offline long enough. The server cleans up the environment, and the terminal running claude remote-control then prints a line that ends the environment was cleaned up on the server and can't be resumed. Run claude remote-control for a fresh environment, and pick up any uncommitted work from the worktrees it lists.

The second is running for about 30 days. Before Claude Code 2.1.267 (9 September 2026), claude remote-control exited and dropped every attached session when its server credential expired about 30 days after it started. Since then it re-registers and keeps going. That fix is in the changelog, and the docs pages don't mention it.

Anthropic does not publish how long "offline long enough" is, and it publishes no overall limit on how long a Remote Control session can run or sit idle beyond the figures above. The numbers here are what the docs state. None of them is a guarantee.

The "Can’t reach your computer" message

Anthropic does not publish this message in its docs. It shows up in user-filed issues on the anthropics/claude-code GitHub repository. Every report we found comes from the Claude Desktop app (for example issue 90172 and issue 90877). Here's the text as people report it:

Can’t reach your computer

It may be asleep or offline. This session will reconnect when it’s back.

The details view reportedly shows a status code, Remote Control host unreachable (computer_unreachable). Some reports quote the message with straight apostrophes instead of the typographic ones above.

What it means. The Claude app cannot reach the machine that hosts the session. The message offers the ordinary causes, asleep or offline, and says the session will reconnect. That matches the docs: the session goes offline until the machine is back or the claude process is started again.

What to do, in order:

• Step 1. Wake the machine and check that it has a network connection. If it was asleep, Claude Code reconnects by itself once the machine is back online. If it was awake but offline for roughly 10 minutes in server mode, claude remote-control has exited, so go to Step 2.
• Step 2. Check that Claude Code is still running there. If the host is the Claude Desktop app, reopen the app on that computer, then open the session again. If it's a terminal that was closed, start it again: claude remote-control in the same directory for server mode, or claude --continue for a session you started with --remote-control or /remote-control.
• Step 3. If the server stopped more than about four hours ago, or the machine was offline long enough to have its environment cleaned up, start a new session with claude remote-control.
• Step 4. Still stuck, with the machine awake and online and Claude Code running? Treat it as a reported bug, since it isn't documented behaviour. Users report three things: a session card that shows as live with no process behind it (issue 90877), a Windows reinstall that regenerates the device identity and orphans sessions (issue 88692), and the Desktop app restarting itself to install an update, with the running sessions not coming back afterwards (issue 90172). Anthropic has not confirmed any of these. If you started the session from the command line, run /remote-control to reconnect. Otherwise start a new session.

The "Claude Code on the computer running this session is offline" message

This one is not in Anthropic's docs either. Users report it in claude.ai/code in a browser and in the Desktop app, sometimes under a heading that begins Remote Control disconnected (issue 91839 and issue 95577). The body, as people report it:

Claude Code on the computer running this session is offline. If that computer is asleep or lost its connection, this session reconnects when it’s back. If Claude Code was closed, start it there again and reopen this conversation.

Reports quote the opening and closing sentences, and the middle sentence appears in a search query and in a newer report. The search query is lowercase, so the capital letters in that sentence are our reading. We could not check it against a live session, so read it as the message people describe, and don't treat it as a screenshot. A variant that names the machine turned up in a report filed on 30 September 2026: Claude Code on <machine> is offline. If <machine> is asleep or lost its connection, this session reconnects when it’s back... (issue 98310).

What it means. It covers two different conditions, and the message tells you which fix applies:

• The computer is asleep or lost its connection. Wake it or restore the network. After a sleep, Claude Code reconnects by itself. In server mode a network outage of roughly 10 minutes exits claude remote-control, so check that it's still running and start it again if not.
• Claude Code was closed. Start Claude Code there again with the commands from the section above, then reopen the conversation in claude.ai/code or the Claude app.

Anthropic documents a similar state for projects, under the heading Lost contact with your folder. The session there stopped responding, usually because the computer went to sleep or the Desktop app or claude remote-control quit. The fix is the same: wake the computer, then restart the app or run claude remote-control again in the same folder (Projects docs). Lost contact with your folder is a different message from the one quoted above, so the docs still don't cover the offline message itself.

One more report matters if you run server mode unattended. Issue 98310 describes a claude remote-control host that ran as a systemd user service. After a session was archived and then unarchived from claude.ai, the host showed the machine-name version of the message, even though it was still connected and polling. It's user-filed, and Anthropic hasn't confirmed it. That's our reading of one report, and it isn't documented advice: if the host looks healthy and the banner says offline, check the session's archive state before you restart anything.

Keep the host online: your options

Remote Control has one requirement that no setting removes: a machine that is on, with claude running. So it comes down to where that machine lives, and whether it keeps going with your laptop shut:

• Your laptop, as it comes: no. Sleep pauses the session, and Remote Control reconnects when the laptop wakes. The Claude Desktop app has a setting, Keep this computer awake for Remote Control, under Settings, then Claude Code. A closed lid is a separate problem, so see the keep-awake command builder.
• A desktop, home server or VPS: yes, while it stays on and online. It keeps going only while the machine, its network and the claude process stay up. Run claude remote-control inside tmux. Anthropic's docs say that to keep a session running on a remote machine after you disconnect from SSH, you start it inside tmux or screen. You own the sleep settings, the login, the updates and the restarts.
• Anthropic's cloud sessions: yes, until they go idle. This isn't Remote Control. Anthropic says the session keeps running after you close your laptop. After a period of inactivity it stops and the machine is reclaimed.
• A Hivra computer. See the last section.

When Anthropic's cloud is the better choice. Say you only use Claude Code, your work lives in GitHub repositories, you don't need local files, tools or MCP servers, and the laptop will be closed. Cloud sessions are the simplest answer. Anthropic's mobile docs say the same about a computer that will be off: use cloud sessions or a project, because Remote Control drives your own machine. Cloud sessions are available on Pro, Max and Team plans, and on Enterprise with premium or Chat + Claude Code seats. They share rate limits with your other Claude and Claude Code usage, carry no separate compute charge, and aren't available to Zero Data Retention organizations.

When Remote Control is the better choice. You're in the middle of a task on a machine with its own files, MCP servers and settings, and you want to steer it from your phone while that machine stays awake. Anthropic's rule of thumb matches that: use Remote Control to keep going with local work from another device, and a cloud session to start something with no local setup.

Set up an always-on machine for Remote Control

Here's a checklist built from Anthropic's docs for a machine nobody sits at. The tmux step is Anthropic's own advice. We haven't run this whole sequence on a server ourselves, so check each step against the current docs.

• Step 1. Pick a machine that doesn't sleep. A desktop, a home server or a VPS. On a Mac, the keep-awake command builder writes the command for your setup.
• Step 2. Sign in through claude.ai. Run claude, then /login. If your browser shows a code instead of redirecting back, paste it at the Paste code here if prompted prompt. Anthropic says this is common in SSH sessions and containers.
• Step 3. Clear out anything that blocks Remote Control. This prints variable names only, never their values:

If ANTHROPIC_API_KEY, ANTHROPIC_AUTH_TOKEN or CLAUDE_CODE_OAUTH_TOKEN shows up, remove it where it's set. Same for a CLAUDE_CODE_USE_* provider variable such as CLAUDE_CODE_USE_BEDROCK, for ANTHROPIC_UNIX_SOCKET, for ANTHROPIC_BASE_URL when it points at a host other than api.anthropic.com, and for CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC or DISABLE_GROWTHBOOK. That check covers your shell only. The same variables can be set in the env block of a settings file, so look there too, and check for an apiKeyHelper while you're in there.

• Step 4. Start it inside tmux and answer the one-time prompts. Run this in the project directory. tmux is a real terminal, so the first-run questions show up there: answer y to Enable Remote Control? (y/n) and to Trust <directory>? [y/N]. The tmux cheat sheet has the keys on one page.

Press the spacebar for the QR code and scan it with your phone. Claude Code 2.1.284 (28 September 2026) added the trust question. Older versions don't ask: an untrusted directory exits with the Workspace not trusted error, so run claude in the folder once first and accept its trust dialog. In an untrusted directory the command also exits with that error whenever standard input or output isn't a terminal. So the first run can't come from a service or a cron job.

• Step 5. Expect it to need a restart now and then. A network outage of roughly 10 minutes exits server mode. A Ctrl+C can be undone within about four hours, and a crashed session in server mode is served again when you message it. Before Claude Code 2.1.267, server mode also exited at about 30 days, so run claude update on a long-lived host. And the login itself expires. Claude Code warns at startup when your claude.ai login is within three days of expiring. Anthropic says a Remote Control session that outlives the login stops making progress until you sign in again (authentication docs). Run /login to renew.

If you want the process to come back by itself after one of those exits, a service manager or a restart loop is the usual tool. That's our suggestion. Anthropic's docs name only tmux and screen. We haven't tested a supervised setup, so there's no unit file here.

Security and limits, as documented

What Anthropic's docs say, with no claims of our own added. Your session makes outbound HTTPS requests and never opens inbound ports. It registers with the Anthropic API and polls for work, and traffic goes through the API over TLS, using several short-lived credentials that each cover a single purpose. Claude runs on your machine. There's no cloud VM and no sandbox by default, so to sandbox the sessions a server starts, turn on sandboxing in a settings file.

• Where the transcript lives. While Remote Control is connected, the session transcript, including your messages, Claude's responses and tool activity, is stored on Anthropic's servers and retained under its Data usage policy. Execution and filesystem access stay on your machine.
• Permission modes from the app. Remote Control sessions offer Manual, Accept edits and Plan. Bypass permissions and Auto can't be selected for them. Server mode also takes --permission-mode at startup.
• Switches for you and for admins. The disableRemoteControl setting turns it off. Trusted Devices, a beta that's off by default, asks for an enrolled device and a sign-in no more than 18 hours old before you can view or steer sessions. On Pro and Max you turn Trusted Devices on yourself, and on Team and Enterprise an Owner does.
• Limits. One remote session per interactive process. Server mode allows 32 sessions at once by default. Terminal-only commands such as /plugin and /resume work only from the local command line.

Approving actions from a phone is easy, which is exactly why a destructive one deserves a second look on a small screen. The guide to what each permission mode does covers Bypass and Auto, which the app can't pick. For the wider risks of leaving an agent running while you're away, see is it safe to leave an AI agent running unattended.

When not to use Remote Control

Use a cloud session or a project if the machine will be off or asleep. Anthropic's own mobile docs say so.

API keys aren't supported, and neither are Amazon Bedrock, Google Cloud's Agent Platform, Microsoft Foundry or a gateway. Running Claude Code on your own key? See BYO API key: what it means.

An organization on Zero Data Retention or a HIPAA configuration is out too. Remote Control can't be enabled there.

Want work to start from an event or a schedule with nobody at the keyboard? Look at what else Anthropic offers. It lists Channels (research preview) for events from Telegram, Discord or CI, and scheduled tasks for recurring work. Its own table lists Remote Control for steering in-progress work from another device.

And if you want the agent off your laptop for good, you need hosting. See the 24/7 guide and why agents die when the terminal closes.

Hivra computers and Remote Control

Hivra gives a coding agent a computer of its own, separate from your laptop. On a paid Hivra plan the computer stays on and keeps your files, sessions and login. A Claude Code or Codex run you start inside tmux in the computer's Terminal tab keeps going after you close the laptop. On Claude Code, work you send through Telegram, connected in the agent's Telegram tab, runs on the computer, not in your browser.

Remote Control has not been tested on a Hivra computer, so we can't say it works there. The keep-running statement above covers tmux and Telegram. It says nothing about Remote Control. Hivra's Telegram tab is Hivra's own connection, separate from Anthropic's Channels feature. If you try Remote Control on a Hivra computer, treat it as untested and check the requirements above first.

Paid plans start at $9.99 a month for 2 vCPU and 4 GB of RAM, stay on without being paused for inactivity, and come with a 7-day money-back guarantee on card payments. The $19.99 plan is 4 vCPU and 8 GB of RAM. Details are on the pricing page, and you can launch Claude Code from the Claude Code agent page. Hivra is independent and is not affiliated with Anthropic or OpenAI.