Claude Code on mobile: code from your phone

There's no Claude Code app you install on your phone. But you can absolutely drive a coding session from it. Here's what that actually looks like and when it's useful.

First, what you're not getting

I should be upfront: Claude Code is a terminal tool. It runs in a shell. Your phone doesn't have that, and Anthropic hasn't built a standalone mobile version. What they have built is a set of ways to control a session that runs somewhere else. Your phone becomes the steering wheel, but the engine is either on your computer or in the cloud.

This distinction matters because it determines what Claude can actually reach. Can it touch your local files? Your database? Your MCP servers? Depends entirely on where the session is running.

Remote Control: your machine, your phone

Remote Control is the most interesting option if you already have Claude Code set up on your computer. You start a session before you leave your desk:

This gives you a QR code and a session link. Scan it with the Claude app on your phone (or open it in claude.ai/code), and you're connected. The session keeps running on your machine. Your working tree, your CLAUDE.md, your local MCP servers, your database: all reachable. Your phone is just the window.

I find this genuinely useful in a specific situation: you've got something running, you walk away, and then you need to poke it. Maybe a test failed and you want to tell Claude to investigate. Maybe you want to check on a long build. You're not retyping your whole project context from scratch; you're picking up a conversation that already has everything loaded.

If your network drops or your laptop briefly sleeps, it reconnects when things come back. It's not fragile. But it does require your machine to be on and awake, which is the obvious constraint. Requires Claude Code v2.1.51+ and a Pro or Max plan.

Claude Code on the Web: a cloud sandbox

The second option is to open claude.ai/code in your phone's browser. This spins up a session in Anthropic's cloud. No connection to your local machine at all. It runs in an isolated sandbox, and you bring code into it (usually by pointing it at a GitHub repo).

This is good for self-contained tasks. You want to prototype something, explore a public repo, or write code that doesn't depend on your specific local setup. It persists in the cloud and you can come back to it from any device.

The tradeoff is clear: no access to your local files, no local MCP servers, no local databases. If the bug you're chasing requires hitting your dev database or reading a config file on your machine, this won't help. Use Remote Control for that.

How to pick between them

The rule of thumb is simple. If the task needs to touch anything on your machine (local files, local services, MCP tools), use Remote Control. If it's a clean task against a repo that lives on GitHub and doesn't need local state, the Web works fine.

In practice, I reach for Remote Control most of the time, because most real work involves local context. The Web sandbox is better for quick throwaway experiments or reviewing code you don't have checked out locally.

Channels and Dispatch: fire and forget

Remote Control and the Web both give you a live session you watch and steer in real time. But sometimes you just want to send a task and get a notification when it's done. That's what Channels and Dispatch are for, and unlike the first two, these are generally available (not just preview).

Channels let Claude Code listen for messages from Telegram, Discord, or iMessage. You set up a channel on your machine, and then you can text a task from your phone: "run the test suite and tell me what broke." Claude runs it in the background and replies in the same chat thread.

Dispatch works through the Claude phone app directly. You pair your phone with your desktop app, and then you can send tasks from the phone that spawn sessions on your desktop. Think of it as a dedicated task queue between your phone and your computer.

Both of these are "fire a task and check later" rather than "sit and watch." They're good for kicking off builds, running test suites, or asking Claude to investigate something while you're busy with other things.

What this actually looks like day to day

Here are the situations where I actually reach for my phone instead of waiting to get back to my computer:

A CI build fails and you get the notification on your phone. You open the Remote Control session and ask Claude to look at the failing test, see the output streaming on your screen, and tell it to push a fix. Total time: a few minutes on the train.

Someone posts a pull request and you want Claude to review it. Open the Web at claude.ai/code, point it at the PR, and read through its analysis on your phone. You don't need local state for a code review.

You remember something you forgot to do before leaving the office. Send a Dispatch task from the Claude app: "add the error handling to the upload endpoint and commit it." Your desktop picks it up and runs it. You check the result later.

None of this replaces sitting down at a real keyboard for serious work. Typing on a phone is still slow, and complex debugging sessions need a bigger screen. But for quick interventions, status checks, and kicking things off, it removes the "I have to be at my desk" constraint.

The rough edges

Remote Control and Claude Code on the Web are both labeled "research preview" for good reason. They work, but you'll bump into limitations. Remote Control needs your machine awake and connected. The Web sandbox doesn't have your local tooling. Phone screens are small and terminal output is dense.

Channels and Dispatch are more polished (GA), but they require initial setup, and the fire-and-forget model means you don't see what's happening until it's done. Sometimes that's fine. Sometimes you want to steer.

My advice: set up Remote Control once, see if it fits your workflow, and add Channels later if you find yourself wanting background tasks. Don't try to adopt all four at once.