Claude Remote Control: Run Claude Code From Your Phone
· 5 min read
You kick off a long refactor in your terminal, close the laptop, and head out — then realize the agent is waiting on a question you can't answer until you're back at your desk. Claude Remote Control fixes exactly that. Released by Anthropic in February 2026, it's effectively a mobile version of Claude Code: start a task on your machine and keep full control of it from your phone, tablet, or the web. Here's what it is, how the security model works, and who can use it.
What Is Claude Remote Control?
Remote Control is a synchronization layer that bridges your local Claude Code session with the Claude mobile app and web interface. You begin a complex task in your terminal as usual, then continue monitoring and directing it from another device — decoupling the AI agent from the physical workstation.
Crucially, the work still runs on your computer. Your phone or browser becomes a "remote window" into the session that's executing locally — not a copy of it.
How It Works
The architecture is designed so you never expose your machine to the open internet:
- You start a session. You run the Remote Control command in your local Claude Code terminal.
- Your machine connects outbound. Your desktop opens an outbound connection to Anthropic's API (which also serves the Claude models powering Claude Code). No inbound ports are opened, and nothing on your machine is exposed to the public web.
- Your devices become remote windows. When you open the session URL or the Claude app, you view and command the process that is still running on your computer.
- Only messages and results travel. Your files and MCP servers never leave your machine — only chat messages and tool results flow through the encrypted bridge. (New to MCP? See What Is MCP?)
- It reconnects automatically. If your laptop sleeps or the network drops, the session stays alive in the background and reconnects when the host machine is back online.
This replaces the old workarounds — SSH from mobile, port forwarding, VPN gymnastics — with a native streaming connection that requires none of them.
Why It Matters
- Untether from your desk. Approve plans, answer the agent's questions, and redirect work from anywhere.
- No risky exposure. Because the connection is outbound-only, you don't open inbound ports or expose your machine.
- Your code stays local. Files and MCP servers remain on your machine; only conversation and tool output cross the bridge.
- Resilient sessions. Auto-reconnect means a closed lid or flaky network won't kill a long-running task.
Who Can Use It
Remote Control launched as a research preview for subscribers on Anthropic's Claude Max tier, with access for Claude Pro users expected to follow. During the initial phase it's positioned as a power-user feature and was notably absent from Team and Enterprise plans. Because it's an evolving preview, availability and tier requirements change — check Anthropic's current documentation before relying on it.
Security Considerations
The outbound-only design is a deliberate, strong security choice: there's no listening port for an attacker to find. That said, a remote-controllable coding agent is still a powerful capability, and the usual agent-safety rules apply:
- Protect the session link. Treat the session URL like a credential.
- Keep humans on consequential actions. Review plans before approving destructive or irreversible steps.
- Mind prompt injection. Any agent that reads untrusted content (issues, web pages, dependencies) can be manipulated — see our deep dive on AI agent security and prompt injection.
- Use deterministic tools for sensitive, exact tasks. Decoding a token or hashing a value is safer in a local, offline tool than routed through any agent — e.g. JWT Decoder or Hash Generator.
Remote Control vs Computer Use vs Cowork
These Anthropic features are easy to confuse:
| Feature | What it does |
|---|---|
| Remote Control | Drive a local Claude Code session from your phone/web |
| Computer Use | Claude controls a screen/UI by clicking and typing |
| Claude Cowork | A desktop agent that works with your files and apps to deliver finished knowledge work |
Remote Control isn't Claude taking over a computer — it's you taking your existing terminal session with you.
Frequently Asked Questions
What is Claude Remote Control?
It's a feature that lets you start a Claude Code task in your local terminal and continue controlling it from the Claude mobile app or web — effectively a mobile version of Claude Code. The work keeps running on your own machine.
Does Remote Control expose my computer to the internet?
No. Your machine makes an outbound connection to Anthropic's API; no inbound ports are opened. Your files and MCP servers stay local, and only chat messages and tool results pass through the encrypted bridge.
Do my files leave my machine?
No. Files and MCP servers remain on your computer. The remote device is just a window into the session running locally — only messages and tool results are transmitted.
Who has access to Remote Control?
It launched as a research preview for Claude Max subscribers, with Claude Pro access expected to follow. It was initially absent from Team and Enterprise plans. Check Anthropic's current docs, as preview availability changes.
Is Remote Control the same as Computer Use?
No. Computer Use is Claude operating a UI by clicking and typing. Remote Control simply lets you monitor and steer a Claude Code session from another device while it runs on your own machine.
Conclusion
Claude Remote Control removes the last tether between you and a running coding agent: start a task at your desk, steer it from your phone, and never expose your machine to do it. With an outbound-only connection, local files, and auto-reconnect, it's a thoughtfully designed bridge for power users who delegate real work to Claude Code. Pair it with sound agent-security habits and your trusted local tools, and you get mobility without giving up control.
Sources: Claude Code Docs, VentureBeat, Help Net Security.