Claude Code on Your Phone: What Actually Works

"Claude Code mobile" is really three different things wearing one app icon. Understanding which one you are reaching for changes what you can actually do, so this guide starts there before getting into SSH clients and notifications.

What "Claude Code on your phone" actually means

There is no separate Claude Code mobile app. The Claude app for iOS and Android is a client, not a place where code runs on its own; a Code tab inside it gives you access to three different underlying things, and the app looks almost identical for all three:

• Cloud sessions running on Anthropic-managed infrastructure, the same sessions reachable at claude.ai/code on desktop.
• Remote Control, which connects your phone to a Claude Code process running on your own computer.
• Dispatch, which lets you message a task to the Desktop app on your computer and have it decide how to run it, on plans that include the feature.

The distinction that matters most day to day is the same one that shows up everywhere in Claude Code's remote options: does the work run on Anthropic's cloud, or on a machine you control. Our guide to running Claude Code remotely goes deeper on that split if you want the full picture beyond mobile specifically.

Cloud sessions from your phone

A cloud session started from the web, the terminal (with the --cloud flag), or directly from the phone runs on Anthropic's own infrastructure, so it keeps going after you put the phone down. This is the best fit for a task on a GitHub-hosted repo that should keep running unattended, for reviewing a diff on the go, or for approving a pull request from wherever you are. Exact plan eligibility, rate limits, and repository requirements are set by Anthropic and can change, so check their current documentation before depending on it for something time-sensitive.

Remote Control: your own machine, driven from your phone

Remote Control keeps execution on your computer and turns your phone into a window into that session: your filesystem, your tools, and your MCP servers are all still there, because nothing actually moved. You start it with claude remote-control (or --remote-control, or /remote-control from an existing session), scan a QR code or open the session from the app, and steer from there. You can send images or files from your phone and Claude Code pulls them into the session as a real file reference, the same as if you had typed a path in the terminal.

The catch is the one you would expect: your computer has to stay on and reachable, because Remote Control is not doing the work, it is watching work your machine is already doing. A brief sleep or network drop reconnects automatically; a longer outage ends the session and you start a fresh one.

SSH clients: the vendor-agnostic option

Independent of any of the above, a plain SSH client on your phone can reattach a tmux or screen session running on any box you control, exactly the way it works from a laptop. Several long-established mobile SSH apps exist for iOS and Android with features built for a small screen, like Mosh support that survives a network change more gracefully than plain SSH. This route works with a box you are already running Claude Code on for other reasons, and it does not depend on any particular vendor's app or feature availability.

The honest limitation is typing. A phone keyboard makes a long, detailed prompt genuinely tedious to compose. Built-in dictation on both iOS and Android works fine into a terminal app, since it is just inserting text, and it is a real mitigation if you find yourself typing long instructions on the move.

Push notifications and alerts

Remote Control's built-in push notifications cover the common case well: turn on Push when Claude decides and Push when actions required from /config, and Claude pushes to your phone when a long task finishes or it needs a decision from you. That covers the session itself asking for your attention.

It does not cover the box going away entirely. If the machine running the session loses power, crashes, or drops off the network, there is nothing left running to send that notification. A separate heartbeat, or dead man's switch, that pings you if the box or the agent goes silent closes that gap, which is exactly the kind of monitoring an always-on setup needs on top of whatever notification path the client you are using provides.

What works well on mobile, and what does not

• Works well: monitoring progress, approving or reviewing a plan, answering a yes-or-no question, a short steering prompt, glancing at a diff, getting pinged when something needs you.
• Does not work well: composing a long, detailed prompt, reviewing a large diff carefully, working across several files side by side, or a sustained session where the phone is your only device.

The realistic framing is that a phone is for steering and monitoring sessions that were started, and mostly finished, somewhere else, not a substitute for a laptop as your primary way of writing code.

Security on mobile

• Lock the app. Turn on biometric or passcode lock for the Claude app and any SSH client, since a phone is more likely to be lost or picked up than a laptop.
• Be careful on public networks. Avoid entering an SSH key passphrase or sensitive credentials over public Wi-Fi; a mesh VPN such as Tailscale reaches your own box without exposing SSH to the open internet at all, which matters even more from a mobile network you do not control.
• Know your organization's device policy. Team and Enterprise plans can require enrolled, recently authenticated devices before Remote Control sessions can be viewed or steered at all; if you hit an unexpected access prompt, that is likely why.

Frequently asked questions