Every way to run Claude Code on a phone
There are now six realistic ways to reach Claude Code from an iPhone, iPad or Android device, and three of them are free. Here is the full set, what each one actually requires, and how to pick.
Two years ago the answer was "SSH in and squint". It is now a real field with first-party options, and the right pick depends on three questions: is your code on GitHub, do you own a machine that stays on, and do you need the agent to work unattended.
We make one of the things on this list, and we have tried to write the rest as if we didn't. Where a free option is better for your situation, it says so.
1. Claude Code on the web (cloud sessions)
Free with Pro/Max · needs GitHub · nothing of yours stays on
Pick a repo and branch in the Claude app's Code tab, describe a task, and it runs on Anthropic-managed infrastructure. It keeps going after you lock the phone, syncs across devices, and can watch a pull request and fix CI failures as they arrive.
Choose it when your code is on GitHub and the task is self-contained. Skip it when the work needs your database, your private network, or a repository that is not on GitHub.
2. Remote Control
Free with Pro/Max · your machine must stay on
Run claude remote-control on your computer (or /remote-control in an open session), scan the QR code, and drive that session from the app. Execution and filesystem access stay local. Push notifications work while it is active.
Choose it when you have a desktop that is usually on and want to steer real local work from the sofa. Skip it when the machine in question is a laptop that goes into a bag. Note also that Bypass permissions is not selectable from the app.
3. Dispatch
Pro/Max only · needs the Desktop app running
Message a task and let Dispatch decide how to run it, via the Desktop app on your computer. Per Anthropic's docs it requires a Pro or Max plan and is not available on Team or Enterprise.
Choose it when you want to lob a task over and not think about where it executes. Skip it when your computer is off.
4. An SSH client and tmux
Free · works with anything · rough on a small screen
Termius, Blink Shell, Prompt or a-Shell, attached to a tmux session on a server. You get the complete terminal, so every slash command and TUI panel behaves exactly as on desktop, and nothing is missing.
Choose it when you also want a real shell on that machine, or you rely on commands no dedicated client wraps. Skip it when reading wrapped diffs and hunting for Esc on a soft keyboard is the thing stopping you from using it — the detail is in this comparison.
5. Third-party agent clients
Free tiers available · agent runs on your machine · session relays via a backend
A category rather than one product. Happy Coder is open source, free, end-to-end encrypted, and covers Claude Code and Codex. Omnara (YC S25) is an agent control plane with a free tier, a web dashboard, parallel agents, an Apple Watch client and voice. Others in the same space include Paseo and CodeRemote.
These generally run a wrapper CLI next to your agent and sync the session through their infrastructure, which is what makes dashboards and multi-device sync possible.
Choose one when you want a free option, run several kinds of agent, or want a browser dashboard alongside the phone. Look closer when you care about how the session gets to your phone and what sits in between — see Maude vs Omnara for how that architecture differs.
6. Maude
Paid · direct SSH · built for an always-on server at full permissions
Connects straight from the phone to any Linux host over SSH — no relay, no account with us — and renders the session as an app rather than a terminal: markdown, collapsible tool cards, native sheets for /model, /mcp and /compact, push when a task finishes. Sessions live in a server-side daemon, so closing the app is a detach rather than a disconnection, and it runs Claude Code at full permissions on the assumption the box is disposable.
Choose it when your work lives on a server you own, you need unattended runs without approval prompts, or your code cannot be checked out onto third-party infrastructure. Skip it when a free first-party option already covers you — which, for a lot of people, it does.
All six, side by side
| Option | Cost | Where code runs | Machine must stay on | Needs GitHub |
|---|---|---|---|---|
| Cloud sessions | Included | Anthropic cloud | No | Yes |
| Remote Control | Included | Your computer | Yes | No |
| Dispatch | Pro/Max only | Your computer | Yes | No |
| SSH + tmux | Free | Any host | Server, not your laptop | No |
| Third-party clients | Free tiers | Your machine | Usually yes | No |
| Maude | Paid | Any host | No, if it's a VPS | No |
First-party details verified against Anthropic's documentation in July 2026; third-party details from each project's public site the same month. This space changes fast — check current sources before committing.
Code on GitHub, want zero setup? Cloud sessions.
Always-on desktop, happy to approve steps? Remote Control.
Want it free and don't mind a terminal? SSH and tmux.
Multiple agents, want a dashboard? A third-party control plane.
Own server, unattended, full permissions, private code? That's the gap Maude fills.
What changed in 2026
The important shift is that Anthropic filled in the easy cases itself. Cloud sessions landed for GitHub-shaped work, and Remote Control arrived in February for driving your own machine. Anyone whose question is simply "can I check on Claude Code from my phone" now has a good free answer built into the app they already have.
What is left for everyone else is the awkward middle: code that cannot go to a third party, environments that cannot be reproduced in a provisioned sandbox, no machine that stays awake, and tasks that must run without stopping to ask. That is a smaller field than it was — and it is worth knowing which side of it you are on before you pay for anything.
