Table of Contents
- What is Claude Code Remote Control, and how does it actually work?
- How do you turn on Remote Control and pair it with the Claude app?
- Which plans, Claude Code versions, and devices support Remote Control?
- Does your computer have to stay on, and what happens if it sleeps?
- Can you approve file edits and shell commands from the phone, and what if you lose signal?
- Is Remote Control safe, and what can a paired phone execute?
- When should you use SSH plus tmux or a third-party app instead of the built-in feature?
- Why does the session not show up in the Claude app, and how do you fix it?
- How did we compare these options?
- How do you get started with Agent 37?
- What do people also ask?
Do not index
Claude Code Remote Control is Anthropic's built-in way to pair a Claude Code session that is already running on your own machine with the Claude iOS, Android, or web app, so you can read output, send prompts, and approve tool calls from your phone. The catch is that the session dies the moment that machine sleeps, so we recommend running it on an Agent 37 persistent instance, where Claude Code runs in an isolated container on the same always-on hosting that serves OpenClaw and Hermes from $3.99 per month. Agent 37 is not a replacement for Remote Control and not a phone client; it is the always-on computer the paired session lives on, and your Anthropic credentials go directly to Anthropic.
Key takeaways
- Remote Control is a window, not a cloud session: the agent keeps executing on the host machine with its repo, environment variables, MCP servers, and tools, and the Claude app only sends prompts and approvals.
- Setup is three commands: update Claude Code, start it inside tmux, run claude remote-control (or /remote-control in a live session), then scan the printed QR code from the Claude app.
- The host must stay awake: if the laptop sleeps, the terminal closes without tmux, or the network drops, the session goes offline and the phone cannot restart it.
- You cannot launch new sessions from the phone: Remote Control resumes a session you started at the keyboard; starting fresh work needs SSH, a wrapper like CouchShip or PandanPC, or a browser terminal on an always-on box.
- Agent 37 solves the always-on problem: isolated containers from $3.99 per month, a full browser TTY terminal, a file browser, a live Linux desktop via noVNC, and automatic runtime updates and security patches.

What is Claude Code Remote Control, and how does it actually work?
Claude Code Remote Control exposes a Claude Code session that is already running on your computer to the Claude mobile or web app. The agent keeps executing on your machine with your repo, environment variables, MCP servers, and tools, while the app acts as a remote screen and keyboard. Nothing moves to a cloud sandbox.
That direction matters, because Anthropic ships three related features that get confused with each other. Claude Code on the web runs the agent in an Anthropic-managed cloud sandbox against a repository cloned from GitHub, with no access to your local files. Session teleport pulls a web session down to your terminal with a flag on the claude command. Remote Control goes the other way: the work stays local and the phone becomes a viewer and approver.
The practical consequence is that anything your local session can already do, the remote view can do too. Local MCP servers, a .env file in the project, a database on localhost, and a Docker daemon are all reachable, which is why the host machine and its permission mode deserve more attention than the pairing step itself. Anthropic documents the current command names and flags in the Claude Code section of the Claude developer docs, which is the place to check if a command below has been renamed.
How do you turn on Remote Control and pair it with the Claude app?
Update Claude Code, open a terminal inside your project, start a tmux session, then run claude remote-control or type /remote-control inside a session that is already running. The terminal prints a QR code and a link; scan it from the Claude app on iOS or Android, or open the link on claude.ai, and the session appears on your phone.
Here is the command-level walkthrough we use on an Agent 37 instance and on a Mac mini alike:
- Update first: run claude update, then claude --version, and make sure you are signed in with claude login using your Claude account rather than an API key.
- Change into the repository you want the agent to work in, then start a detachable terminal: tmux new -s rc. This is the step most people skip and most people regret.
- Start the session with Remote Control enabled: claude remote-control (the short alias is claude rc). If a session is already open, type /remote-control or /rc at the prompt instead.
- The terminal prints a QR code and a pairing URL. Open the Claude app on iOS or Android, go to the Code area, and scan the code; on the web, open claude.ai and pick the session from the list of active sessions tied to your account.
- Detach the terminal with Ctrl-b then d. The process keeps running inside tmux, and your phone now shows the live transcript, accepts new prompts, and surfaces permission requests.
One decision rule: give each remote session a tmux window named after the task. When you come back to the desk, tmux attach -t rc drops you straight into the same transcript the phone was showing, with nothing lost.
Which plans, Claude Code versions, and devices support Remote Control?
Remote Control shipped as a research preview for Claude Pro and Max subscribers who sign in to Claude Code with their Claude account, and it works from the Claude iOS app, the Android app, and the claude.ai web app. API-key billing, Team, and Enterprise availability has changed, so confirm your plan on Anthropic's pricing page before relying on it.
The version requirement is simpler than it looks: run claude update and the feature is either present or the command tells you it is not. Older installs pinned by a package manager are the usual reason the subcommand is missing. Plan tiers and current prices are listed on Anthropic's pricing page, and the Pro, Max 5x, and Max 20x tiers are the ones that carry the Claude Code subscription allowance.
A remote session is still the same session, so it draws down the same usage windows as a desk session. If you plan to leave an agent running on a task for an afternoon and poll it from your phone, read how the 5-hour session window and weekly cap in Claude Code interact, because a long autonomous run can hit the rolling window while you are away from the keyboard.

Does your computer have to stay on, and what happens if it sleeps?
Yes. Remote Control is a window into a process on your machine, so if the laptop sleeps, the terminal closes, or the network drops, the session ends and the phone shows it as offline. Keep the host awake, run Claude Code inside tmux, or move the session to an always-on instance such as Agent 37.
This is the limitation that shapes every other decision. A phone cannot wake the machine, cannot restart Claude Code, and cannot start a new session, so the whole feature is only as reliable as the computer underneath it. On macOS, caffeinate -i claude remote-control keeps the Mac awake for the life of the process, and on a Linux desktop, masking sleep.target with systemctl stops suspend entirely. Closing the lid still wins on most laptops unless you also change the lid policy.
The usual fixes are an old laptop, a Mac mini, or a cheap VPS, and all three work. The trade-off is that you now own patching, uptime, and security for a box that runs an agent with your credentials. This is the job we built Agent 37 for: a persistent Claude Code hosting instance in an isolated container, with a full TTY terminal, a visual file browser, and a live Linux desktop via noVNC in the browser, so you can start the tmux session from any device, pair it with the Claude app, and close the tab. Automatic runtime updates and security patches are included, and the managed OpenClaw and Hermes plans on the same platform start at $3.99 per month.
Can you approve file edits and shell commands from the phone, and what if you lose signal?
Permission prompts for file edits, shell commands, and MCP tool calls appear in the Claude app the same way they appear in the terminal, and tapping allow or deny on the phone resolves them. If your phone loses signal, the session keeps running on the host; pending prompts wait, and the app reconnects when the network returns.
The permission mode you launch with decides how often your phone buzzes. Claude Code's modes are default, acceptEdits, plan, and bypassPermissions. For remote work we use default or acceptEdits: file edits inside the repo go through, and anything that touches the shell, git push, or a network call still asks. Plan mode is the right choice when you want to read a proposed approach on the train and approve execution later.
Never run bypassPermissions on a machine that holds real credentials just to reduce prompts. The better fix is to narrow what the agent can do in the first place, which is what a layered setup of constraint files and hooks does; our guide to guardrails for autonomous coding agents walks through a configuration you can copy. One more operational rule: a dropped connection on the phone is not a failed task, so resist the urge to resend the same prompt the moment the app reconnects. The host already has it.
Is Remote Control safe, and what can a paired phone execute?
Pairing goes through your Claude account, so only a device signed in as you can see the session, and the QR code or link is a pairing credential tied to that account rather than an open inbound port on your machine.
Because the host process makes an outbound connection to Anthropic, Remote Control works inside Docker and dev containers and on a VPS exactly as it does on a laptop; there is no port to forward and no firewall rule to open. That also means the container or VM boundary is your real security layer. An agent running as your user on a laptop can read .env files, SSH keys, and cloud credentials, and approval prompts do not change that fact; our post on how coding agents read your repo and how to keep them in bounds explains why a dedicated workspace beats a prompt.
Treat the phone like any other remote access path. NIST's SP 800-46 guide to remote access and BYOD security recommends multi-factor authentication and device protections for every remote channel, so enable MFA on the Claude account and a device lock on the phone before pairing. On Agent 37, each instance is an isolated container and your API keys go directly to Anthropic or OpenAI, so a compromised session on one instance has no path to another, and we never see the credentials.
When should you use SSH plus tmux or a third-party app instead of the built-in feature?
Use built-in Remote Control when you only need to monitor and approve a session you started at your desk. Use SSH plus tmux when you need to start new sessions, run git, or fix the machine itself. Use a wrapper like CouchShip, PandanPC, or Remote Claude when you want multi-session management or voice control from the phone.
The DIY route is an SSH client such as Termius or Blink on the phone, Tailscale so the host is reachable without a public IP, and tmux so the session survives the client disconnecting. It is the only option that lets you launch a brand-new Claude Code session, switch repositories, or reboot a stuck process from the phone, at the cost of typing in a terminal on a six-inch screen.
Third-party wrappers fill the gap Anthropic left. CouchShip and PandanPC focus on launching and managing multiple sessions from a mobile interface, Remote Claude adds voice control through iMessage and Siri, and claude-remote-kit is an open-source starting point if you want to build your own bridge. None of them removes the always-on machine requirement, and each adds a layer that can read your transcript, so check the vendor's site for current pricing and data handling before you install one.
Option | Best for | Pricing or format | Standout limit |
Agent 37 instance + Remote Control | Sessions that must outlive your laptop and stay patched | Managed hosting from $3.99/mo, bring your own Anthropic login | Not a phone client; you still pair through the Claude app |
Built-in Remote Control | Watching and approving a session you started at the desk | Included with Pro and Max; confirm tier on Anthropic's pricing page | Cannot start new sessions; dies when the host sleeps |
SSH + tmux + Tailscale | Starting new sessions, git work, and fixing the host | Free tooling; Termius or Blink may charge for pro features | Terminal on a phone; you manage the host's security |
CouchShip, PandanPC, Remote Claude | Multi-session management or voice control from mobile | Varies by vendor; check each site for current pricing | Still needs an always-on machine; adds a third party to the path |

Why does the session not show up in the Claude app, and how do you fix it?
Most failures trace to one of four causes: the phone and terminal are signed in to different Claude accounts, the pairing code expired before you scanned it, the host went to sleep or the terminal closed without tmux, or a VPN such as Tailscale changed the route mid-session. Fix the account first, then restart remote-control inside tmux.
Work down this list in order. Run claude login on the host and compare the email shown with the account in the Claude app; a work account on the phone and a personal account in the terminal is the single most common cause of an empty session list. If the accounts match and the code still fails, generate a fresh one, because the QR code is meant to be scanned promptly rather than saved for later.
If the session appeared and then vanished, the host is the suspect. Check whether the machine slept, whether the terminal window was closed outside tmux, and whether a VPN reconnect gave the host a new route. Reattach with tmux attach, run /remote-control again, and rescan. On an Agent 37 instance this whole loop happens in the browser terminal, so the recovery takes one tab and no trip home to a sleeping laptop.
How did we compare these options?
We compared the four approaches on what each lets you do from a phone, whether it can start a new session, whether it needs an always-on machine, and the security model of each path. For pricing, we used Anthropic's published plan page, the vendors' own sites, and Agent 37's own published rates. We did not run benchmarks or surveys.
How do you get started with Agent 37?
Deploy an Agent 37 instance in one click, open the browser terminal, run claude login and then claude remote-control inside tmux, and scan the code from the Claude app. The instance stays on after you close the tab, your credentials go straight to Anthropic, and managed hosting starts at $3.99 per month.
If you want the session to outlive your laptop today, create your always-on Agent 37 instance and pair it with the Claude app in a few minutes.
What do people also ask?
How do I turn on Remote Control in Claude Code and connect it to the Claude app on my phone?
Run claude update, start Claude Code inside tmux in your project folder, then run claude remote-control (alias claude rc) or type /remote-control in a live session. The terminal prints a QR code and a link; scan it with the Claude iOS or Android app, or open the session on claude.ai with the same account.
Does my computer have to stay on for Claude Code Remote Control, and what happens if it sleeps?
Yes. The agent runs on your machine, so if it sleeps, the terminal closes without tmux, or the network drops, the session goes offline and the phone cannot restart it. Keep the host awake with caffeinate or a sleep policy, or run the session on an always-on Agent 37 instance from $3.99 per month.
Can I start a brand-new Claude Code session from my phone with Remote Control?
No. Remote Control only resumes a session you started at the keyboard. To launch new sessions from a phone you need SSH plus tmux over Tailscale, a wrapper such as CouchShip or PandanPC, or a browser terminal on an always-on instance like Agent 37.
What is the difference between Claude Code Remote Control and Claude Code on the web?
Remote Control keeps the agent on your own machine with your local files, env vars, and MCP servers, and the phone is only a viewer and approver. Claude Code on the web runs in an Anthropic-managed cloud sandbox on a repo cloned from GitHub, with no access to your local environment.
Is Claude Code Remote Control safe to use, and what can a paired phone execute?
Pairing is tied to your Claude account and uses an outbound connection, so no inbound port is opened. The risk is the host: a paired phone can run anything the local user can run within the permission mode you chose, so use default or acceptEdits, enable MFA on the account, and run the session in an isolated container rather than on a laptop holding real credentials.
