Remote control

Control a local Claude Code session from the browser or Claude mobile app; tools still run on your machine.

Remote control lets you step away from the terminal while keeping one continuous Claude Code session. Your laptop or workstation remains the host that actually runs tools, shell, and MCP, while claude.ai/code or the Claude mobile app becomes a second UI attached to the same thread.

Remote Control left research preview in August 2026. Official guide: Remote Control.

What Remote Control is

Remote Control is a control-plane bridge, not a second copy of your repo in the cloud. Reads, edits, builds, and MCP calls execute through the local Claude Code process you started. Messages, responses, and tool activity are still transported to and stored on Anthropic’s servers under its data-usage policy, so “local execution” does not mean the conversation is local-only.

That distinction matters for private networks, licensed CLIs, local databases, and anything that would be awkward or impossible to reproduce in a generic cloud sandbox.

How it works (high level)

Host starts Remote Control: claude remote-control starts server mode without a local interactive session so the phone or browser can choose a directory and create sessions. For a local interactive session that is also remotely controllable, use claude --remote-control / --rc. From an already-running session, /remote-control or /rc connects it.

Outbound coordination: the local process talks to the service over HTTPS so the remote UI can stay in sync. You typically do not open a random inbound port on your LAN for strangers to hit; the security story is closer to “this machine initiates a trusted session” than “I'm running a public server.”

Client attaches: you open the session URL on another device or scan the QR code with the mobile app, sign in as the same Claude.ai identity, and pick the named session when the host shows online.

Shared conversation, local execution: prompts and tool permission prompts appear on both sides; when the agent runs Bash, Edit, or MCP, the paths and side effects are on the host. If you close the host terminal, Remote Control for that session ends.

Throughout, treat the remote device like a machine that can steer and approve work: screen lock, account security, and who is holding the phone still matter.

Compared to Claude Code on the web

Claude Code on the web executes in cloud infrastructure. Remote Control instead lets the browser or phone steer a process on your machine, retaining access to your local repository, private network, tools, and MCP servers.

Practical constraints

Availability and org policy: Remote Control is available on Pro, Max, Team, and Enterprise. Team and Enterprise organizations have it off by default until an Owner enables it.
Auth: Remote Control requires an eligible Claude.ai subscription login. API-key authentication is not supported, and sessions routed through a non-Anthropic ANTHROPIC_BASE_URL gateway cannot use it.
Startup behavior: it activates only when you run one of the commands above unless you enable Remote Control for all sessions in /config (remoteControlAtStartup).
Host lifetime: the local machine must remain online. Ending the server or interactive host disconnects remote control.
Concurrency: an interactive Claude Code process hosts one remote session. claude remote-control server mode can host multiple concurrent sessions, with a default capacity of 32.

Claude Code on the web

Cloud execution instead of remotely steered local execution.

Parallel and scheduled work

Background sessions, worktrees, goals, and routines.