Recipe 017
Leave the Laptop Home: Vibe Coding from Your Phone
Remote Control has quietly become one of our most-used Claude Code features: the session keeps running on your machine at home, and your phone becomes the steering wheel. Here's the setup, the honest limits, and the strange little thrill of shipping a fix from a park bench.
Ideas have terrible timing. The fix for last night's bug arrives in the supermarket queue. The perfect next feature shows up on a walk. And the old reality was: note it down, wait until you're back at the desk, hope the spark survives the commute.
That reality quietly ended for us. Lately one of the most-used features in our kitchen is Remote Control — and the first time you steer a coding session from a park bench, phone in one hand, coffee in the other, something clicks that doesn't fully un-click.
Quick signal flare: this one's for you if you already run Claude Code on your own machine with a Claude subscription (setup here if not). It doesn't work with API keys — subscription login only.
The kitchen stays home. You carry the order window.
Here's the mental model, because it's what confused us at first: Remote Control does not move your session to the cloud. Your machine at home keeps cooking — same project folder, same files, same MCP servers, same everything. What travels is the conversation: the Claude mobile app (or claude.ai/code in any browser) becomes a second window into the same session.
So you're not getting a stripped-down phone version of your agent. You're getting your actual kitchen — the one with your knives in it — controllable from wherever you are.
Setup: one command before you leave
In your running Claude Code session, type:
/remote-control
That's it for the common case. Your session shows up in the Claude app's session list with a little computer icon and a green dot. Tap it, and you're in the same conversation you left at your desk. (There's also a QR-code pairing flow, and a claude remote-control server mode that can host several sessions — but the slash command is the whole story for day one.)
Three requirements worth checking before you trust it:
- Claude Code v2.1.51 or newer —
claude --versiontells you. - Logged in via
/loginwith your claude.ai account — subscription plans (Pro/Max/Team) only; API keys don't work. Team/Enterprise admins have to switch the feature on first. - The laptop stays on and the process stays running. Close the terminal, and the session ends. This is the price of "it's really your machine."
What it's actually like
The workflow that made this a daily habit for us:
Start something long, then leave. Kick off a refactor or a batch of tests at the desk, type /remote-control, pocket the phone, go live your life. The agent keeps working — your machine is doing the cooking, remember.
Check in from anywhere. Open the app, read what happened, answer the question the agent's been politely waiting with. Course-correct in two sentences. This is the part that changed how we work: long tasks used to mean sitting there. Now they mean checking in.
Send what you see. Since v2.1.202 you can send photos and files from the phone — snap a whiteboard sketch or an error on someone else's screen, and it lands on your machine for the agent to work with.
And yes, the part nobody puts in the docs: the first time you merge a real fix from a waiting room, you will feel like the only person in that room with a superpower. That feeling is not the point. But it doesn't hurt, either.
The honest notes
- Your machine is the single point of truth. Laptop sleeps briefly? It reconnects. Network gone for more than ~10 minutes? The session times out. Terminal closed? Game over. Plan accordingly (power settings matter).
- Not everything works from the phone. Text commands like
/compact,/usage,/contextare fine; interactive pickers (/resume,/plugin) are desk-only. Permission approvals work, with limits — if you run with approvals on, expect to handle the occasional prompt when you're back. - It's a research preview. Rough edges exist; the official docs are current and short — worth the three minutes.
- Don't confuse it with Claude Code on the web. claude.ai/code can also run cloud sessions on Anthropic's machines — great for kicking off tasks with zero local setup, but that's a rented kitchen without your local files and MCP servers. Remote Control is the opposite deal: your kitchen, your tools, controlled remotely.
Order up
Tonight, before you leave your desk, paste this into Claude Code:
I want to try Remote Control before I leave my desk.
1. Run `claude --version` in the terminal and tell me if I meet the
v2.1.51 minimum. Confirm whether I'm logged in with a claude.ai
account (not an API key), and if not, tell me exactly what to run.
2. Walk me through enabling Remote Control for this session, and how
to find it in the Claude mobile app afterwards.
3. Then start a genuinely useful task that will take a while — e.g.
[write tests for X / document module Y / audit Z for dead code].
4. Standing rule while I'm remote: whenever I check in, start your
reply with a two-line status — what's done, what you need from me.
Then actually leave. Check your phone from the sofa, the gym, the bus. The task will have moved without you — and the next time an idea shows up with terrible timing, you'll be able to do something about it on the spot.
Where's the strangest place you've shipped from? Tell us at @AgentKitchenHQ.