Use Claude Code from your phone: 4 setups that keep agents working

You can use Claude Code from your phone today: the Claude app for iOS and Android connects to sessions on your own computer (Remote Control), to cloud sessions run by Anthropic, or you can SSH into a server where Claude Code runs. This guide sets up each option, explains which one survives a closed laptop, and shows what kind of work suits a phone.

Can you use Claude Code from your phone?

Yes. Your phone becomes a remote control for an agent that runs somewhere else. Claude Code never executes on the phone itself. The real question is where the agent runs: on your own computer, on Anthropic's cloud, or on a server you control. That choice decides whether the work continues when your laptop lid closes.

According to the Claude Code mobile docs, there is no separate Claude Code app. Sessions live in the Code tab of the regular Claude app for iOS and Android. From there you can reach:

• Remote Control: a Claude Code session running on your computer.
• Cloud sessions: a session running on cloud infrastructure, Anthropic-managed by default.
• Projects: a conversation where Claude coordinates several parallel threads of work.
• Dispatch: a task you message to the Claude Desktop app on your computer (Pro or Max).

Outside the Claude app, the classic route still works: SSH into a machine where Claude Code runs inside tmux, using any phone terminal app.

Which phone setup should you choose?

Choose by where the code needs to live and whether your computer will be on. This table covers the four practical setups.

Remote Control is the quickest to try if you are at your desk now. Cloud sessions are the easiest way to keep work going overnight. A server gives you the most control and works with Codex, Gemini CLI and other agents too.

How do I set up Claude Code Remote Control?

Run claude remote-control in your project, scan the QR code with your phone, and the session appears in the Claude app. The agent keeps running locally, with your filesystem, MCP servers and tools, while you type from the phone.

Per the Remote Control docs, it works on Pro, Max, Team and Enterprise plans (on Team and Enterprise an Owner must enable it first). API keys are not supported, and it does not work through Bedrock, Google Cloud, Foundry or a custom ANTHROPIC_BASE_URL.

1. Install the Claude app on your phone and sign in with the same account you use for Claude Code. Running /mobile inside Claude Code shows a QR code that opens the right app store.

2. On your computer, open your project and start a Remote Control server:

   cd ~/code/my-appclaude remote-control --name "my-app"

3. Accept the one-time confirmation. Press the spacebar to show a QR code and scan it, or open the Claude app, tap Code and pick the session.

4. Type your task from the phone. You can attach photos (Claude sees them directly) and files (downloaded to your machine and passed in as @ references).

There are two other ways to start it:

Server mode has a useful flag for phone work: --spawn worktree gives each new session its own git worktree, so two tasks you start from the couch do not edit the same files.

How do I start a cloud session from the Claude app?

Open the Code tab, pick a GitHub repository and branch, describe the task, and submit. The session runs on cloud infrastructure, so you can lock the phone and the work continues.

Cloud sessions are available on Pro, Max and Team plans, and for Enterprise users with eligible seats. They need access to your code on GitHub, which you grant either by installing the Claude GitHub App during web onboarding or by running /web-setup in your terminal to share your local gh token.

Each session runs in a cloud environment that controls network access, environment variables and setup scripts. Spend ten minutes on that environment before relying on it from a phone: if npm install or your test database fails at startup, the agent cannot verify its own work and you will get a weaker result.

Cloud sessions also move between devices. Start one from your phone and it is waiting at your desk; start one on your laptop with claude --cloud and check it from the app. You can also ask Claude to watch a pull request and fix CI failures or review comments as they arrive.

The trade-off: each cloud session starts from your repository and environment config, not from your laptop. Local-only tools, private services on your home network, and files that are not in git are not there.

How do I run Claude Code over SSH from a phone?

Run Claude Code inside tmux on an always-on Linux server, connect over a private network, and attach from a phone SSH app. This setup works with any coding agent, any plan or API key, and any phone.

1. Get a server that stays on: a small VPS or a spare machine at home. Install Claude Code with the native installer and sign in once.

2. Put the server and your phone on a private network so SSH is not exposed to the internet. With Tailscale, for example, you install it on both devices, and running tailscale set --ssh on the server lets Tailscale handle SSH authentication, per the Tailscale SSH docs.

3. Start a named tmux session and run the agent inside it:

   tmux new -s agentcd ~/code/my-appclaude

4. Detach with Ctrl-b then d. The agent keeps running.

5. From your phone's SSH app, connect and reattach:

   tmux attach -t agent

A few habits make this bearable on a small screen. Increase the font size in your SSH app and turn the phone sideways to read diffs. Keep prompts short and put the detail in files the agent can read (a task file or your CLAUDE.md). Learn three tmux keys: detach, switch window, and scroll mode (Ctrl-b then [).

The cost of this route is maintenance. You patch the server, manage disk space, keep tools updated and secure the box. Our guide to setting up a remote development server for AI agents covers hardening, users and backups in detail.

Why does Claude Code stop when I close my laptop?

Claude Code is a process on your computer, so it pauses when the computer sleeps, loses network, or shuts down. Remote Control does not change that; it only lets you see and type into the session from elsewhere.

You have three ways around it:

• Keep the computer awake (on macOS, caffeinate -i while a task runs) and accept the battery, heat and "laptop in a bag" risks.
• Move the work to the cloud with cloud sessions, which run independently of your laptop.
• Run the agent on a machine that never sleeps: a VPS, a home server, or a hosted box.

For a full comparison of these options, including headless claude -p runs and scheduled tasks, see how to run Claude Code 24/7.

Where a hosted box fits

A hosted box is a server someone else maintains, where your agents keep running after you close the app. It sits between cloud sessions (no persistent machine) and a self-managed VPS (persistent, but all the admin is yours).

Orkio is one example: it rents an isolated, always-on cloud server where Claude Code and Codex run on your own Claude account, so there is no markup on model usage. You give agents tasks from your phone or the web, close the app, and review the pull requests later. The Starter box (2 GB memory, 1.5 vCPU) is $19 a month, and the phone reaches the box through an end-to-end encrypted relay.

A hosted box makes sense when you want the persistence of a VPS (installed tools, cloned repos, running dev servers) without patching a server, and when you want to steer several agents from a phone app built for that rather than a terminal.

What can you realistically do from a phone?

A phone is good for starting, steering and reviewing work, and poor for anything that needs you to read a lot of code at once. Plan your phone sessions around short inputs and clear outputs.

Tasks that work well from a phone:

• Kicking off a well-scoped task you already thought through.
• Answering the agent's questions and approving permission prompts.
• Asking for a summary of what changed and why.
• Reviewing a pull request description and small diffs.
• Queueing follow-ups: "Now add tests for the edge cases you listed."

Tasks to save for a laptop: debugging a race condition, reviewing a 1,000-line diff, or anything where you need two files side by side.

The quality of a phone session depends on the first message. Compare:

fix the signup bug

Signup fails for emails with a plus sign (user+tag@example.com).Find where we validate emails, fix the check, and add a test casefor plus addresses. Run the auth tests. Open a PR with a shortsummary of the cause. Ask me before changing the database schema.

The second prompt names the symptom, the expected change, how to verify it, and where to stop. That is what lets the agent finish without you watching. Our guide on how to write tasks for AI coding agents has templates you can save as phone snippets.

How do you keep phone-driven agents safe?

Decide the permission mode before you walk away, and get notified when the agent needs you. An agent waiting on an approval you never see wastes hours; an agent with unlimited permissions on your laptop is a security risk.

• Permission mode: start Remote Control with a mode that fits the task, for example claude remote-control --permission-mode acceptEdits for edit-heavy work. Claude Code's permission modes range from default (asks before most actions) to auto (a classifier reviews actions) to bypassPermissions, which the docs restrict to isolated containers and VMs.
• Push notifications: with Remote Control active, run /config in your terminal and enable Push when actions required (permission prompts and questions), Push when Claude decides (for example, a finished task), or both. You can also write "notify me when the tests finish" in your prompt.
• Prompts stay open: permission prompts and Claude's questions are forwarded to the app and wait until you answer.
• Isolation for unattended work: if the agent will run for hours without you, run it on a separate machine or container rather than the laptop that holds your SSH keys and browser sessions.

How to start today

Pick the smallest setup that matches your situation and try one real task tonight.

• Install the Claude app and sign in with your Claude Code account.
• At your desk: run claude remote-control in a project and send one task from the phone.
• Enable push notifications in /config so you know when it needs you.
• For overnight work: connect GitHub and run one cloud session from the Code tab, or set up tmux on a server.
• Write your first phone prompt with a symptom, a change, a check and a stopping point.
• Review the result in the morning and note what the agent needed that it did not have.

If you plan to make this a habit, the overnight AI agent workflow shows how to queue tasks in the evening and review them in ten minutes the next morning.

Frequently asked questions