Run Claude Code on Several Computers, Manage It From One Phone
Run multiple Claude Code sessions across your computers and track them from one list on your phone: see which needs you, reply, approve and start new ones.
The Coder IDE team · · 7 min read
Running one Claude Code session is easy to keep an eye on. Running four, spread across a desktop, a laptop and a home server, is where things slip: one session sat waiting on a permission prompt for an hour, another finished and nobody looked. This post covers why people run agents in parallel, what goes wrong when they do, and a setup that puts every session on every computer into one list on your phone.
Why people run Claude Code in parallel
A single agent spends a lot of time doing things that don't need you: reading files, running tests, waiting on a build. If you only run one, you spend that time watching it. Run several and the waiting overlaps, so more work gets done in the same afternoon.
The common patterns look like this:
- One session per task. A bug fix in one session, a refactor in another, a test backfill in a third.
- Git worktrees. Each session works in its own checkout of the same repository, so two agents never edit the same files on disk. Claude Code's own background sessions do this for you: a dispatched session moves into its own worktree before it edits anything, according to the agent view docs.
- Separate projects. Different projects open in different VS Code windows, each with its own Claude Code chat.
- Separate machines. The heavy build runs on the desktop with the big GPU, a small script fix runs on the laptop.
All of these work. The cost shows up somewhere else.
The real problem: which session needs me?
Marc Nuri describes it well in his post on building an AI coding agent dashboard: at five to ten concurrent sessions he starts forgetting about some of them and finds abandoned work hours or days later. Terminal tabs and tmux keep sessions alive, but they don't tell you which one is blocked, and they don't help at all when the sessions live on different computers.
So the human becomes the bottleneck. Every agent is either working, waiting on a question, waiting on a permission prompt, finished, or stopped with an error. Only some of those states need you, and the expensive part is finding out which.
Parallel agents only pay off if checking on them is cheaper than the time they save. That means one view, sorted by what needs you.
What Claude Code already gives you
Be clear about what you get for free before adding anything.
Claude Code has an agent view of its own, opened with claude agents. It is one terminal screen for your background sessions, grouped by state: working, needs input, idle, completed, failed, stopped. You can peek at a session's output, reply without attaching, and dispatch new sessions into their own worktrees. The docs note that it is a research preview and that each session uses your subscription quota independently.
Anthropic's Remote Control lets you continue local sessions from the Claude app or a browser, and its server mode can host several sessions from one process. It can also push a notification to your phone when a session needs a decision.
If all your sessions run on one computer and you're happy in a terminal, those two may be all you need. Where it gets harder is several computers, sessions started in different places, and wanting to act on a session from a phone without opening a terminal.
One list for every session on every computer
Coder IDE runs the VS Code server on your own computers and lets you open them from a browser or the Android app. Alongside the editor, it shows every Claude Code session on all of your connected computers, whether the session started in a terminal, inside VS Code, or in the background.
The Sessions list is grouped by what needs you, so the sessions waiting on an approval or a reply sit at the top and the ones that are busy or finished sit below. Each session shows its steps with the command or file involved, and long runs fold into one line such as "Ran 14 commands · edited 3 files", so you can read what happened without scrolling through every tool call.
You don't have to keep the list open. You get a push notification when a session:
- needs your approval for a command or edit,
- is waiting for your reply,
- stops unexpectedly.
Tap the notification, read the full request, and approve or deny it from your phone or a browser. Approving something you haven't read is how agents end up doing things you didn't want, so the full request is always there before you decide. The approvals post goes deeper on what to approve from a phone and what to save for a bigger screen.
Start and reply to sessions from your phone
Triage is half the job. The other half is keeping work moving without walking back to the desk.
From your phone or browser you can reply to any session, which covers most "waiting for you" states: Claude asked which approach you prefer, or wants to know whether to keep going after a failing test. You can also start a new session in any of your projects on any connected computer. The computer picks it up even while you're away, and Claude Code keeps working when your phone sleeps, you switch apps, or your connection drops.
A practical rhythm for parallel work looks like this:
- Start two or three sessions at the desk, each with a clear task and a clear definition of done.
- Leave. Let notifications tell you when something needs a decision.
- Answer short questions and routine approvals from your phone.
- When a session finishes, review its changes file by file, comment on the lines you want changed, and send all the comments back to Claude in one message.
- Merge at the desk once you've seen the result running.
Reviewing on a small screen has its own tricks, covered in reviewing Claude Code's diffs on your phone.
Agent view: a focused layout, one tap from the full IDE
Sometimes you want more than a list but less than a full IDE on a phone screen. Coder IDE's Agent view is a focused AI layout inside the editor: the Claude Code chat front and centre, with new chats opening as tabs right inside it. One tap switches back to the full IDE when you want the file tree, the terminal or the Ports panel. (Same name as Claude Code's terminal claude agents screen, different thing: this one lives in the editor.)
When a review comment is quicker to fix by hand, Open in VS Code takes you straight to the file and line you were looking at.
Keep parallel agents from stepping on each other
More agents means more chances for two of them to touch the same thing. A few habits prevent most collisions:
| Risk | What to do |
|---|---|
| Two sessions editing the same files | Give each session its own worktree or branch, or keep them in different projects |
| One dev server port fought over by two tasks | Check the Ports panel, which lists servers started in any terminal on the computer, before starting another |
| A session you forgot about | Let the grouped Sessions list and notifications surface it instead of relying on memory |
| Two big tasks slowing one machine | Put the heavy one on another computer; they all show up in the same list |
| Losing track of which window is which | Use a separate project per folder and name it |
Coder IDE lets you open each folder as its own project on the same machine and switch between them, so each project gets its own window and its own Claude Code chat. If you turn on auto-suspend for idle editors, it waits for a running Claude Code task to finish before shutting anything down.
Also keep an eye on cost. Each parallel session spends your Claude quota on its own, so five agents burn through it five times as fast as one.
Wrap-up
Parallel agents are worth it when you can see, in one glance, which of them needs you. On a single machine, Claude Code's own agent view covers a lot. Once your sessions spread across several computers and you want to act on them from a phone, Coder IDE gives you one list grouped by what needs you, notifications for approvals and stalls, replies and new sessions from anywhere, and the real VS Code one tap away. If you're starting out, setting up an async workflow is a good next read.
Start free. The free plan covers one computer, and Pro adds unlimited computers when you outgrow it.
