Claude Code Remote Control connects claude.ai/code or the Claude app for iOS and Android to a Claude Code session running on your own machine, so you can start a task at your desk and keep steering it from a phone or another browser (Remote Control docs). You turn it on with claude remote-control, claude --remote-control or /remote-control, and Claude keeps running locally the whole time.
What runs where
Remote Control is a window into a local session, not a copy of it. Your filesystem, MCP servers, tools and project configuration stay available, the conversation stays in sync across every connected device, and typing @ from the phone or browser autocompletes file paths from your local project. Cloud sessions use the same claude.ai/code interface but execute on cloud infrastructure, Anthropic-managed by default; the docs compare Remote Control with Dispatch, Channels and scheduled tasks in one table. The cost of running locally: your computer has to stay on and the claude process has to keep running.
For a setup like mine, that split matters. The rules for this blog live on my machine: a project CLAUDE.md with positioning and copy rules, and scripts/content_lint.py, which fails a run on em dashes, links to removed pages and over-long titles. A session steered from a phone is still that local session, with the same files and scripts in reach.
Three ways to start it
| Command | What you get |
|---|---|
claude remote-control | Server mode: the terminal waits for remote connections and can serve several sessions |
claude --remote-control or claude --rc | A normal interactive session you can also type into locally |
/remote-control or /rc | Run inside an open session; carries over the current conversation |
The Claude Desktop app and the VS Code extension use /remote-control as well (start a session).
Set it up
Check 1: Is your login eligible?
claude, then /login, and sign in through claude.ai.ANTHROPIC_API_KEY or ANTHROPIC_AUTH_TOKEN, remove it from your shell or your settings' env block, then run claude auth login. API keys are not supported.ANTHROPIC_BASE_URL pointed at a gateway or proxy, and with CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC or DISABLE_GROWTHBOOK set.Check 2: Does the session start?
claude remote-control in your project directory, or claude --remote-control "My Project" for a named interactive session.Enable Remote Control? (y/n), then shows a session URL; press spacebar for a QR code.y to Trust <directory>? [y/N]. If server mode refuses a global flag such as --settings, drop the flag: it is not carried over to the sessions the server starts (server mode).Check 3: Can your phone find the session?
myhost-graceful-unicorn, which your first prompt replaces./mobile in Claude Code for a QR code to the right app store (connect from another device).Check 4: Is the connection live?
/rc active indicator in an interactive session, or run /remote-control again to open the status panel./remote-control to reconnect, unless the reason says the session changed elsewhere, such as another connection taking it over.Check 5: Will your phone tell you when Claude needs you?
/config enable Push when actions required, Push when Claude decides, or both./config shows No mobile registered, open the Claude app so it refreshes its push token.Check 6: Will it survive you walking away?
tmux or screen. On a laptop, leave the claude process running.claude remote-control in the same directory, or claude remote-control --continue for only the session it started with (resume). For claude --remote-control or /remote-control, use claude --continue or claude --resume.Turn it on for every session
Run /config and set Enable Remote Control for all sessions to true, or set remoteControlAtStartup to true in your user ~/.claude/settings.json (auto-connect). A checked-in .claude/settings.json can turn it off for a repository but not on. Auto-connect signs in with your own claude.ai account, and the sessions it starts appear only in your account's Claude apps.
What the docs say about security
Your machine makes outbound HTTPS requests only and never opens inbound ports. The session registers with the Anthropic API and polls for work, and all traffic travels through that API over TLS, using short-lived credentials, each scoped to one purpose (connection and security).
While Remote Control is connected, the transcript, including your messages, Claude's responses and tool activity, is stored on Anthropic servers and retained under the data usage policy. Execution and file access stay local. The disableRemoteControl setting turns the feature off entirely, organizations with Zero Data Retention can't enable it, and an organization with a HIPAA configuration gets a policy error instead of a session.
For a tighter gate, Trusted Devices, in beta, requires every browser, phone or desktop app that views or steers a session to be enrolled, plus a sign-in no more than 18 hours old, refreshed with Face ID, Touch ID, Windows Hello or a passkey (Trusted Devices). On Pro and Max you turn on Require trusted devices yourself; on Team and Enterprise an Owner does.
Limits to plan around
From the docs' limitations and server mode flags:
- One remote session per interactive process. Server mode runs several, up to 32 at once by default (
--capacity). - By default, server-mode sessions share the current directory and can conflict on the same files;
--spawn worktreegives each new session its own git worktree. - If the machine is awake but offline, server mode gives up after roughly 10 minutes and exits. An interactive session keeps retrying until the network returns.
- Terminal-only commands such as
/pluginand/resumework only from the local CLI./compact,/clear,/context, and/modelor/effortwith the value as an argument work from mobile and web. - Permission prompts and
AskUserQuestionquestions wait until you answer. Other forwarded dialogs close after five minutes by default and take their no-action default;dialogExpirychanges that.
Who still approves
On this blog, agents never commit or push on their own; I approve every commit and push. Remote Control does not change who decides: permission prompts reach the phone and stay open until answered (limitations). For multi-agent work, a connected device shows the subagents and workflows already running in the background, and stopping one there stops it on the machine (connect from another device). One run on October 2, 2026 used 32 agents to write 13 manuals, tighten 5 drafts and render-check 20 pages, which is the kind of job a second screen is for.
Where this fits
Remote Control lets you steer a long run from anywhere; it does not decide how the run is split. That structure, one coordinator handing out work and an independent verifier checking it, is in a coordinator, workers and a verifier.
Tags
Frequently asked questions
Does Claude Code Remote Control work with an API key?
No. It needs a Pro, Max, Team or Enterprise plan signed in through claude.ai. If ANTHROPIC_API_KEY, ANTHROPIC_AUTH_TOKEN or an apiKeyHelper setting is in use, Claude Code names it and refuses; remove it and run claude auth login. A long-lived token from claude setup-token can only make model requests, so it cannot open a session either.
Does Claude Code Remote Control keep working if my laptop goes to sleep?
The session pauses and comes back. If the laptop sleeps or the network drops, Claude Code reconnects automatically when the machine is online again. If you close the terminal or quit the app, the session goes offline until you bring it back; a server stopped with Ctrl+C can be resumed with claude remote-control in the same directory for about four hours.
What is the difference between Claude Code Remote Control and Claude Code on the web?
Where the session runs. Remote Control executes on your machine, so your local files, MCP servers, tools and project configuration stay available. A cloud session runs on cloud infrastructure, Anthropic-managed by default, and suits a task with no local setup or a repo you have not cloned. Both use the claude.ai/code interface.
Can I start a new Claude Code session from my phone?
Yes, in server mode. While claude remote-control runs, you can start sessions from claude.ai/code or the Claude app, up to 32 at once by default. They share the current directory unless you start the server with --spawn worktree, which gives each new session its own git worktree.