Claude Code Remote Control is Anthropic’s way to continue a Claude Code session that is running on your machine from a phone or a browser. An agent room is a shared space where a few people you trust can send work to the coding agent on your machine while you approve what runs. Remote Control is about you on another device. A room is about other people.
If you only ever want to reach your own agent from the couch, Remote Control is the right tool and it is built in. If a co-founder, a teammate or a partner needs to ask your agent for something, you need a room. This post lays out the difference precisely so you can pick, or combine them.
What Remote Control does
Remote Control, released as a research preview in February 2026, turns any browser or the Claude mobile app into a window onto a Claude Code session running locally. You start it with claude remote-control or /remote-control inside a session. The session registers with the Anthropic API, polls for work and streams messages back. Your code, environment variables, MCP connections and project configuration stay on your machine, and the connection is outbound HTTPS only, so nothing opens a port.
The limits are the ones you would expect from a personal remote: it is single-user, so you cannot invite a colleague into the session; each Claude Code instance supports one remote session at a time outside server mode, and only one device can be connected at once; the terminal has to stay open and the machine awake. Availability started with Max subscribers and has been widening since, so check the current docs for your plan.
What an agent room does
An agent room gives the agent an address that other people can reach. The agent is still Claude Code or Codex on your machine, with your subscription and your files. A small daemon polls muti for work over an outbound connection, runs the request locally, and streams every tool call back into the room. The people you invite sign in with a browser and install nothing.
Three things follow. Several people can talk to the same agent, each with their own permission mode. Requests queue in the agent’s inbox while the machine is asleep and run when it is back, because the daemon can be installed as a service. And the four dangerous action types, pushing to a shared branch, deleting, spending money and touching credentials, pause for you no matter who asked. See what is an agent room for the full definition.
Remote Control connects one person to one session. A room connects several people to one agent, with the owner approving.
Side by side
When Remote Control is the better choice
- You are the only person who will ever talk to the agent.
- You want zero extra software and you already have an eligible plan.
- You want to see the session exactly as it looks in your terminal, including the parts a shared timeline would summarise.
- You are happy to keep the terminal open while you are away.
When a room is the better choice
- Someone other than you needs to ask the agent for work: a co-founder rerunning tests, a designer exporting assets, a partner converting a folder of videos.
- You want to say yes or no per request, or mark one person as trusted while another stays read-only.
- You want the risky calls to stop for you even when you are not watching.
- The machine may be asleep when the request arrives and you would rather it queue than fail.
- You want a history you can hand to someone: who asked, what ran, what came back.
Using both
They compose. Remote Control is a good way to drive your own session from a phone while you are the one working. A room is the way to receive work from others. The same machine can run a Remote Control session in one terminal and the muti daemon as a background service, because both use outbound connections and neither needs an open port.
Frequently asked questions
Does muti replace Claude Code Remote Control? No. Remote Control is for you on another device; a room is for other people. Many owners use Remote Control to work and a room to share.
Can I use a room without giving anyone else access? Yes. A room with only you in it is a personal inbox for your agent, and the phone workflow works the same way.
Does either one move my code to the cloud? No. Both keep the agent, the repositories and the credentials on your machine. Remote Control relays your session through the Anthropic API; muti relays timeline events through its room service. Neither uploads your working tree.
Which runtimes work? Remote Control is a Claude Code feature. A muti room works with Claude Code and Codex today, and the daemon detects other runtimes installed on the machine.
If the question you are actually asking is “how do I let someone else use my agent”, give it a room.
Published · Updated