Claude Code Remote Control, Explained

The oldest problem with Claude Code is that it works in a terminal and you have a life. You kick off a long task, and now you're tethered: the session is on your laptop, the laptop is on your desk, and the only way to know whether Claude finished — or stalled on a permission prompt twenty minutes ago — is to go look.

Claude Code Remote Control is Anthropic's answer to the tether. It lets you connect to a live, running Claude Code session from the Claude mobile app or a browser — see what it's doing, answer its questions, and steer it — while everything keeps executing locally on your machine. Here's how it works, how to use it well, and where it still leaves a gap that notifications fill.

What Remote Control actually is

Remote Control is not "Claude Code in the cloud." The session stays exactly where it always was: a local process, in your project directory, with your files, your configuration, and your MCP servers. What changes is the front end. When you enable Remote Control, your local Claude Code process registers with Anthropic's API over an outbound HTTPS connection, and the Claude app on your phone (or a browser tab) becomes a second screen for that same session.

Two details matter for the security-minded. First, the connection is outbound only — your machine opens no inbound ports and isn't reachable from the internet. Second, your source code doesn't move to the cloud; the remote client sees the conversation and tool activity, while execution and files remain local.

The workflow it unlocks

The canonical use looks like this: you brief Claude Code on a chunky task at your desk — a refactor, a test-suite cleanup, a long migration — and then you leave. From the sofa, the coffee run, or the school pickup line, you open the Claude app, tap into the session, and check on it. If Claude hit one of its questions ("should I also update the fixtures?"), you answer from your phone and it keeps moving instead of idling until you return.

Recent Claude Code releases have made the remote view considerably more useful: foreground subagent activity now streams live to remote clients, so you can watch tool calls and results as they happen rather than staring at a spinner. Combined with background tasks and subagents, it turns a phone into a legitimate monitoring station for work that used to require sitting at the terminal.

The fine print: limits worth knowing

Remote Control is deliberately conservative, and its constraints shape how you should use it:

• The terminal must stay alive. Your machine has to remain on and online, with the Claude Code process running. Close the laptop lid without preventing sleep, and the session — and your remote connection to it — goes with it.
• One remote client per session. Each Claude Code instance supports a single remote connection at a time; this is a personal second screen, not a team dashboard.
• Reconnection is automatic but not magic. Brief network blips and phone sleep are handled gracefully, provided the underlying local process is still running.
• Permissions still apply. Remote Control doesn't loosen anything — the same approval prompts appear, just now on a device that's in your pocket.

Remote Control vs. notifications: pull vs. push

Here's the gap. Remote Control is a pull mechanism: it answers "let me check on Claude" brilliantly, but it only helps if you think to open the app. It won't tap you on the shoulder. If Claude finishes — or blocks on a question — while your phone sits face-down, nothing happens until you look.

That's the same problem that exists at the desk, one window away from the terminal, and the fix is the same: a push layer that tells you the moment your attention is needed. We've covered the DIY routes before — hooks-based notifications on the desktop and push alerts to your phone via ntfy, Pushover, or Telegram. The two mechanisms compose beautifully: the notification tells you Claude needs you, and Remote Control lets you act on it from wherever you are, no laptop required.

A sane setup for long-running sessions

• Prevent sleep before you walk away. On a Mac, caffeinate (or plugging in and disabling sleep on power) keeps the local process alive while you're out.
• Brief thoroughly, then leave. Remote answering works, but typing detailed course corrections on a phone keyboard doesn't. Front-load the context — our Claude Code tips post covers how — so remote interactions stay short: approve, answer, nudge.
• Add a push notification for the finish line. Let a notifier watch the session so you don't have to poll the app to learn that the task ended ten minutes ago.
• Review the diff at a real screen. Approving a permission prompt from a phone is fine; reviewing a 40-file refactor is not. Save judgment calls for the desk.

The bottom line

Remote Control removes the last excuse for babysitting a terminal. Your session runs at home; your control surface rides in your pocket; and with a notification layer on top, the loop closes completely — Claude works, you live your life, and you find out the instant it's your turn again.