How to use Claude Code from your phone

The Claude app now runs Code on your phone: cloud sessions, Remote Control into your own machine, and chat channels. What runs where, which plans get what, and the one thing every path needs.

The short answer

There’s no separate Claude Code app. The Claude app for iOS and Android has a Code tab, and it drives three different things: cloud sessions that run on Anthropic’s machines, Remote Control sessions that run on your machine, and Dispatch, which picks for you. There are also chat channels that put Claude Code in Telegram, Discord, or iMessage. Every option except the cloud one shares a catch: your own machine has to be on and awake the whole time.

The map: five surfaces, two places code runs

Anthropic’s phone story splits by where the session actually executes (all of the below are official features, most labeled research preview, as of August 2026):

Claude Code on the web: sessions run on an Anthropic-managed cloud VM, cloned from your GitHub repo. Start one at claude.ai/code or from the app, close your phone, come back later: “sessions persist even if you close your browser.” Available on Pro, Max, and Team plans (Enterprise needs the right seat type). There’s no separate compute charge: cloud VMs share your plan’s rate limits.
Remote Control: the phone becomes a window into a session running on your own computer. You run claude remote-control on the machine (it even shows a QR code to open on your phone), and the app steers it. Outbound HTTPS only, no inbound ports. It can push notifications to your phone, including “notify me when the tests finish.”
Dispatch: you message a task in the app and Dispatch decides how to run it. Needs the desktop app running, and a Pro or Max plan.
Channels: Telegram, Discord, and iMessage bridges. A channel is an MCP server that pushes your chat messages into a Claude Code session already running on your machine, and the answer comes back in the same chat.
Routines are not a phone surface but worth knowing while you’re here: saved prompts that run autonomously on cloud infrastructure on a schedule or GitHub trigger. You review the results later, from anywhere.

Cloud sessions: the zero-infrastructure path

If your project lives on GitHub, this is the least setup: start a task from the app’s Code tab, and it runs on Anthropic’s VM against a clone of your repo. The cross-device story is genuinely good: a task you start on your laptop is ready to review from your phone, and vice-versa. When you want the session back in a real terminal, claude --teleport pulls it down. Two caveats: the handoff is one-way from the CLI (you can pull a cloud session into your terminal, but you can’t push a terminal session to the web: only the desktop app can do that), and after a teleport the terminal gets its own copy, so the phone-visible cloud session stops reflecting new work unless you start /remote-control in the local session.

Remote Control: your machine, your files, phone in hand

This is the one most people picturing “Claude Code from my phone” actually want: the session executes on your computer, against your real files and tools, and the phone is the steering wheel. It’s available on every paid plan (admin-enabled on Team/Enterprise), and the transport is sensible: your machine makes outbound requests only and never opens a port.

The fine print is where the reality lives. The local process must keep running: close the laptop lid and the session dies. Anthropic’s own docs tell you to start it inside tmux or screen on a remote machine so it survives your SSH connection dropping. That is the point where “phone access” quietly becomes “you need an always-on computer somewhere.” A desktop that never sleeps works. So does a small server that runs Claude Code in tmux full-time, the pattern developers keep converging on.

Channels: Claude Code in your chat app

Channels (research preview) skip the app entirely: you message a Telegram or Discord bot (or, on a Mac, plain iMessage) and the message lands in the Claude Code session you already have open. Setup is a plugin install, a bot token, and a pairing code with an allowlist, then restarting Claude with the channel flag:

/plugin install telegram@claude-plugins-official
claude --channels plugin:telegram@claude-plugins-official

Permission prompts can be relayed to the chat so you can approve or deny remotely.

The load-bearing sentence in the docs: “Events only arrive while the session is open, so for an always-on setup you run Claude in a background process or persistent terminal.” Same conclusion as Remote Control: the chat interface is only as alive as the machine underneath it. If the idea of an agent that’s always there in your chat app appeals more than a coding tool does, that’s the shape OpenClaw was built around from the start.

The DIY lane: SSH clients and third-party remotes

Before any of this existed, people drove Claude Code from phones with an SSH client, and that path still works fine: an always-on box, the agent in tmux, and an app like Termius on the phone. Termius’s own guidance for AI agents on mobile is exactly that pattern: run the agent inside tmux so it survives disconnects, and auto-attach on connect. There are also third-party control surfaces: T3 Code (MIT-licensed, from the t3.gg team) puts Claude Code, Codex, and others behind one free mobile app driving the agents installed on your computer. Third-party means exactly that: your session transits their tunnel or their app’s design decisions, so weigh trust accordingly.

What every path has in common

Strip the branding and there are only two architectures: rent a session on someone’s cloud (Anthropic’s VMs, capped by your plan, GitHub-shaped), or keep a machine of your own awake and reach into it. The second gives you persistence, your real environment, and no ceiling but the hardware, at the cost of being your own operator: a laptop that must never sleep, or an always-on computer that belongs to your agent so your phone is only ever a window, never the thing holding the session up.