Claude Code Remote Control

Problem: You start a long Claude Code run at your desk, then you need to leave. A build is compiling. The couch is calling. Your choices used to be ugly: stay glued to the workstation, or kill the session.

Quick Win: Launch a remote-controllable session from the terminal:

claude remote-control

This command hands your local session to Anthropic's API and prints a URL plus a QR code. Point a phone, tablet, or any browser at it. The full local environment stays put: filesystem, MCP servers, tools, project config. Nothing ships to the cloud.

What Remote Control Actually Is

Remote Control is a research preview feature (February 2026) that links your local Claude Code terminal to claude.ai/code, the Claude iOS app, and the Claude Android app. Think of it as a sync layer, not a lift-and-shift. The session keeps running on your machine the whole time.

Compare this to Claude Code on the web. Web sessions live on Anthropic-managed cloud boxes. Remote Control sessions live on your machine. The web and mobile UIs are windows into that local session.

Why does this matter? Your local rig is irreplaceable. CLAUDE.md, custom skills, file system access, MCP integrations: all of it stays wired up. A cloud session boots fresh every time. Remote Control carries the whole environment with it.

Prefer chat apps over claude.ai? Claude Code Channels lets you steer sessions from Telegram and Discord instead.

Setting Up Remote Control

Requirements

• Claude Code installed: New to Claude Code? Run the platform installer first.
• Subscription: Pro or Max only. API keys are not accepted.
• Authentication: Run claude, then /login to sign in via claude.ai.
• Workspace trust: Run claude in your project folder at least once and accept the trust prompt.

Starting a Fresh Session

Go to your project directory and run:

claude remote-control

The process keeps running in your terminal. A session URL appears, and pressing spacebar reveals a QR code for fast phone handoff. While the session is live, the terminal streams connection status and tool activity.

Flags you can pass:

• --verbose for detailed connection and session logging
• --sandbox / --no-sandbox to toggle filesystem and network isolation

Converting an Existing Session

Already deep in a conversation and want to switch to mobile? Use the slash command:

/remote-control

Or the shorter version:

/rc

This flips your live session into a remote-controllable one and keeps the full history. Tip: run /rename first so the session has a clear label when you hunt for it across devices.

Always-On Remote Control

Want every new session to be remote-ready? Run /config inside Claude Code and flip Enable Remote Control for all sessions to true.

Connecting From Another Device

Three ways in, once a session is live:

1. Open the session URL directly in any browser at claude.ai/code
2. Scan the QR code in your terminal (spacebar toggles it) to jump into the Claude mobile app
3. Pick it from the session list in claude.ai/code or the Claude app. Remote Control sessions are marked with a computer icon and a green dot when online

Messages stay in sync across every connected device. Type from the terminal, the browser, and the phone. The conversation keeps up.

No Claude app yet? Run /mobile inside Claude Code for a download QR code (iOS or Android).

How It Works Under the Hood

The security story is tidy. Your local Claude Code process makes outbound HTTPS requests only. No inbound ports open on your machine. When you start Remote Control, the process registers with the Anthropic API and polls for work. Connect from another device, and the server relays messages between the client and your local session over a streaming link.

Every byte travels through the Anthropic API over TLS, the same transport security every Claude Code session uses. Several short-lived credentials are in play, each scoped to one purpose and expiring on its own clock.

Plain version: your files and MCP servers stay on your machine. Only chat messages and tool results cross the encrypted bridge.

Remote Control vs OpenClaw

If you've watched the OpenClaw story, the pitch sounds familiar. Controlling your computer from your phone is one of OpenClaw's marquee features, and it helped carry the project to 199K GitHub stars.

Remote Control is Anthropic's first-party answer. Here is the head-to-head on this one capability:

The core split: Remote Control is a secure, purpose-built bridge for dev work. OpenClaw is a general life assistant that happens to do remote device control alongside a long list of other tricks. Different tools, different jobs.

For devs who just want to keep a coding session going from their phone, Remote Control is the cleaner path. No exposed ports. No WebSocket CVEs. No self-hosting tax.

Current Limitations

This is a research preview. Expect rough edges:

• One session at a time: Each Claude Code instance accepts a single remote connection
• Terminal must stay open: Close the terminal or stop the claude process and the session dies
• 10-minute network timeout: If your machine can't reach the network for about 10 minutes, the session expires. Run claude remote-control again to spin up a new one
• Permission approval still required: Remote control does not bypass the permission prompts. The --dangerously-skip-permissions flag reportedly does not work with Remote Control yet
• Availability: Pro and Max plans only for now. Not on Team or Enterprise plans

Simon Willison flagged a couple more rough edges in early testing: crashing sessions emit "mysterious API errors" instead of clean termination messages, and restarting the process forces a brand-new session rather than reconnecting to the old one.

Practical Workflows

Here is where Remote Control earns its keep:

The "Walk Away" Pattern

Kick off a big multi-agent task at your desk, then monitor and steer from your phone while you grab coffee. The agents keep running locally with full tool access.

The "Couch Review" Pattern

Queue code reviews or test runs at your workstation, then read results and approve actions from the couch. It pairs well with async workflows where tasks run on their own and you drop in at checkpoints.

The "Multi-Device" Pattern

Start in the terminal for heavy coding, move to the browser on your laptop for lighter work, then use your phone for quick approvals. The conversation stays perfectly lined up across all three.

Pair It With Worktrees

For maximum flexibility, combine Remote Control with git worktrees. Fire up an isolated worktree session, turn on remote control, and drive it from anywhere. Your main branch stays clean while you direct the isolated session remotely.

What's Next

Remote Control shipped alongside Cowork (scheduled tasks), which tells you where Anthropic is heading: a persistent, always-on Claude Code. The current limits, especially the terminal-must-stay-open rule and the one-session-per-instance cap, point to the obvious next move: a fully cloud-hosted Remote Control that runs without your laptop being awake.

Today, Remote Control turns your phone into a window to your terminal. That by itself reshapes how you use long-running Claude Code sessions. No more sitting at your desk watching a build finish or waiting on an agent to wrap up.

Run claude remote-control and see where it fits. If you already run multi-agent setups or long autonomous jobs, the ability to steer from your phone is a real quality-of-life upgrade.