Claude Code Remote Control Is Here, and It Changes the 'Away From Desk' Problem

Anthropic announced Remote Control for Claude Code today. The pitch: start a coding session in your terminal, walk away from your desk, and keep going from your phone. No re-explaining context. No copy-pasting. The session just follows you.

It’s available right now as a research preview for Max subscribers, with Pro support coming soon. You start it with claude remote-control from your terminal, or type /rc inside an existing session.

How It Actually Works

Remote Control doesn’t move your session to the cloud. Your local Claude Code process stays running on your machine the whole time. Your filesystem, your MCP servers, your project config, your tools: all local. The phone (or tablet, or second browser) is just a window into that local session.

When you launch it, Claude Code registers with the Anthropic API and polls for work over outbound HTTPS. No inbound ports open on your machine. Traffic routes through Anthropic’s API over TLS, using short-lived credentials scoped to single purposes.

The connection flow is straightforward:

Run claude remote-control in your project directory
A URL and QR code appear in your terminal
Scan the QR code with your phone, or open the URL in any browser
You’re in. Same conversation, same context, same local environment.

You can send messages from your terminal, your phone, and a browser tab all at the same time. Everything stays in sync. If your laptop sleeps or your Wi-Fi drops, the session reconnects automatically when connectivity returns. A sustained outage of roughly 10 minutes or more will time out the session, but you can restart with another claude remote-control.

There’s also a set-and-forget option: run /config inside Claude Code and flip “Enable Remote Control for all sessions” to true. Every session becomes remotely accessible by default.

The Constraints

A few things to know before you get excited:

One remote session per Claude Code instance. If you’re running multiple instances (say, in separate worktrees), each gets its own isolated remote session.
Your terminal has to stay open. Close the terminal or kill the process and the session ends. This isn’t a background daemon.
Pro and Max plans only. API keys don’t work. You need to be authenticated through claude.ai via /login.
Not available for Team or Enterprise plans yet. Research preview means individual subscribers first.

OpenClaw: The Other Way to Run an Agent From Your Phone

Remote Control isn’t the first answer to “I want to talk to my AI agent from the couch.” OpenClaw, the open-source project created by Peter Steinberger, has been solving this problem from a completely different direction for months.

Where Claude Code’s Remote Control gives you a window into a local coding session, OpenClaw is a full general-purpose agent that uses messaging apps as its primary interface. You talk to it through WhatsApp, Telegram, Slack, Discord, Signal, iMessage, or a dozen other platforms. It runs 24/7 on your own hardware (or a VPS), maintains persistent memory across weeks of conversations, and can read your files, control your browser, manage your smart home, and write code to extend its own capabilities.

The architectural difference matters. Claude Code Remote Control is a coding agent you can reach from your phone. OpenClaw is a life-automation agent that happens to also write code.

OpenClaw’s Mobile Story

OpenClaw’s mobile approach goes deeper than Remote Control’s in some ways. Beyond the messaging app integration, they ship dedicated companion apps through their “Nodes” system:

iOS and Android apps that expose your phone’s camera, location, screen recording, and SMS capabilities to the agent
An Apple Watch app (shipped in the February 19 update) that lets you interact with your agent from your wrist
WebSocket-based pairing that connects companion devices to your central Gateway

Your phone isn’t just a chat interface. It’s a sensor array that the agent can use. OpenClaw can take a photo through your phone camera, read your GPS coordinates, or record your screen, all because the companion app exposes those hardware capabilities to the agent running on your server.

Claude Code Remote Control doesn’t do any of that. It’s purely a conversation and tool-execution interface. Your phone is a terminal emulator, not a peripheral.

The Security Gap

This is where the comparison gets uncomfortable for OpenClaw. Security researcher Wiz found over 135,000 exposed OpenClaw instances online. CVE-2026-25253 was a critical remote code execution vulnerability. The project has improved its security posture since, but the attack surface of a self-hosted, always-on agent with filesystem access and 15+ messaging integrations is fundamentally larger than a session that only makes outbound HTTPS calls through a single provider’s API.

Claude Code Remote Control inherits the security model of Claude Code itself: sandboxed execution, granular permissions, no inbound ports, TLS everywhere, short-lived credentials. It’s a smaller surface area by design, because it’s doing less.

Cost

OpenClaw is free and open-source. You pay for the LLM API calls (OpenAI, Anthropic, local models, whatever you connect) and your hosting. A DigitalOcean droplet runs about $6/month. The agent itself costs nothing.

Claude Code Remote Control requires a Claude Pro ($20/month) or Max ($100-200/month) subscription. That gets you the model, the infrastructure, and the remote connectivity. No self-hosting, no API key management, no server maintenance.

Different Tools for Different Problems

These two solutions aren’t really competing with each other. They’re answering different questions.

Claude Code Remote Control answers: “I’m in the middle of debugging something and I need to step away from my desk. Can I keep going from my phone?” It’s a coding workflow tool. The session is temporary. The context is your codebase. The interaction model is the same as sitting at your terminal.

OpenClaw answers: “I want a persistent AI agent that knows me, runs all the time, and I can reach from any messaging app on any device.” It’s a personal automation platform. The session is permanent. The context is your life. The interaction model is texting a very capable friend.

If you’re a developer who uses Claude Code and sometimes needs to step away mid-session, Remote Control is the obvious choice. It adds exactly one capability (phone access) without changing anything about how Claude Code works.

If you want an always-on agent that handles your calendar, your smart home, your messages, and also writes code when you ask, OpenClaw is the more ambitious tool, with the corresponding complexity and security considerations.

Some people will run both. Claude Code for coding, OpenClaw for everything else. They don’t conflict.

First Impressions

I’ve been using Remote Control for the past few hours. The QR code flow is fast. You scan, you’re in, you’re typing. The sync between terminal and phone is tight enough that I can send a message from my phone and see the tool calls streaming in my terminal simultaneously.

The practical use case that clicked for me: kicking off a long task at my desk, walking to the kitchen, and approving tool permissions from my phone while waiting for coffee. It’s not going to replace sitting at a real keyboard for serious work. But for monitoring, approving, and nudging a session that’s already running, it removes the “chained to my desk” feeling that comes with long agentic coding sessions.

The biggest limitation is the terminal-must-stay-open requirement. If Anthropic eventually lets Remote Control sessions survive a closed terminal (maybe via a background daemon or system service), this becomes significantly more useful. Right now, you’re still tethered to a running process on your machine.

One More Thing

This entire blog post was written using Claude Code Remote Control. I started the session at my desk, walked to another room, and finished it from my phone. The local Claude Code process read the blog directory, checked the frontmatter format of existing posts, and wrote the MDX file, all while I was typing prompts from the Claude mobile app.

That’s the feature working as advertised.

Sources: