Claude Code Remote Control — Drive a Local Session From Your Phone or the Web

Source: Anthropic Remote Control docs (ai-research/claude-code-docs-remote-control-2026-09-29.md); Week 34 what’s-new digest; changelog snapshots 2026-05-01, 05-15, 06-26, 08-07, 08-14; GitHub releases v2.1.224, 232, 238, 251, 257, 260, 273, 281, 283, 284; the launch walkthrough video; talks and courses by Anthropic (London, Tokyo), Sid Benesaria, Chris and Nate, plus the “How the 1%” masterclass; for comparison, the Channels docs, Dispatch tutorial and routines launch post.

Remote Control connects claude.ai/code or the Claude iOS and Android app to a Claude Code session running on your machine. You start a task at your desk and keep steering it from a phone or another browser. Claude keeps running locally, so your filesystem, MCP servers, tools and project configuration stay available; the phone is a window into that session.

Key Takeaways

• Local execution, remote steering. The computer has to stay on and the claude process has to keep running.
• Three ways in: claude remote-control (server mode), claude --remote-control (interactive), or /remote-control inside a running session, including Desktop and VS Code.
• Needs a claude.ai login on Pro, Max, Team or Enterprise; Team and Enterprise need an Owner to switch it on. API keys, Bedrock, Google Cloud’s Agent Platform, Microsoft Foundry and non-Anthropic ANTHROPIC_BASE_URL values are all excluded.
• Transcripts are stored on Anthropic servers while connected. Zero Data Retention organizations can’t enable it.
• Out of research preview since Week 34 (Aug 17–21, 2026), when machines running claude remote-control began showing as device cards in the app’s Code tab.
• Use it for work already running on your machine; for a closed laptop, use cloud sessions or routines.

Start a session

• Server mode, claude remote-control: waits for connections and prints a session URL; press spacebar for a QR code. You don’t type locally. Flags include --name, --spawn worktree and --capacity <N> (default 32).
• Interactive, claude --remote-control "My Project": a normal terminal session that is also reachable remotely.
• From a running session, /remote-control [name]: carries your conversation over. Run it again for the URL, the QR code and a disconnect option.
• First use asks once: Enable Remote Control? (y/n), or an Enable Remote Control dialog for /remote-control.
• Connect by opening the URL, scanning the QR code, or picking the session under Code in the app. A green dot means it’s online.
• Always on: in /config, set Enable Remote Control for all sessions, or put remoteControlAtStartup: true in user or managed settings. Project or local settings can only turn it off.

What you can do from the phone

• Message from terminal, browser or phone interchangeably. Subagent progress stays in sync, and stopping a background task from the device stops it on your machine.
• Attach photos, which Claude sees directly, and files, which download to your machine as @ references.
• Pick the model (v2.1.238+) or the effort level (v2.1.234+) from the device, and it applies on your machine.
• Run /compact, /clear, /context, /usage, /model <name>, /effort <level>, /rename, /mcp, /config key=value (mobile) and /advisor <model> (v2.1.260). /plugin and /resume are terminal-only.
• Get push notifications when a long task finishes or Claude needs a decision. Turn on Push when Claude decides and Push when actions required in /config.

Limits and failure modes

• One remote session per interactive process; server mode serves several. The masterclass presenter thought only one could connect “as far as I’m aware”.
• Closing the terminal, Desktop app or VS Code takes the session offline. On a remote host, use tmux or screen.
• Sleep: the docs say it reconnects when a sleeping laptop comes back online. The launch video said both “I think sleep is okay” and “If your computer falls asleep, the session dies”.
• Network loss: server mode exits after roughly 10 minutes; interactive sessions keep retrying.
• Forwarded dialogs expire: dialogs other than permission prompts and questions close after five minutes (dialogExpiry).

Security and permissions

• Outbound HTTPS only, no inbound ports; TLS through the Anthropic API with short-lived, single-purpose credentials.
• disableRemoteControl turns it off entirely; IT can set it in managed settings.
• Session URL: the launch video warns “anyone could go to this link and then control your local files”, though a second account it tried “was taking a long time to connect”. The docs only say an auto-connected session “grants no one else access”. ^[ambiguous] Keep the URL and QR code private.
• Trusted Devices (beta) requires an enrolled device and a sign-in under 18 hours old, refreshed with Face ID, Touch ID, Windows Hello or a passkey. Owners enable it on Team and Enterprise; Pro and Max users enable it themselves.

Release history

At launch (the video’s demo brief is dated February 25) it was Pro and Max only: “you can’t be on a team or enterprise plan yet”.

Version (date)	Change
v2.1.110 (Apr 15, 2026)	Push notification tool added
v2.1.139 (May 11)	Disabled when ANTHROPIC_API_KEY, apiKeyHelper or ANTHROPIC_AUTH_TOKEN is set
v2.1.222 (Aug 4)	Repo-local settings can no longer turn auto-start on
v2.1.224 (Aug 7)	Compaction progress and /clear reach attached clients; blank “(no content)” after /clear fixed
v2.1.225 (Aug 8)	SendMessage can open conversations with Remote Control sessions on other machines
v2.1.232 (Aug 13)	Keeps reconnecting for about 30 minutes after a network blip
W34, v2.1.234–239 (Aug 17–21)	Device cards in the Code tab; out of research preview
v2.1.238 (Aug 20)	Phone model picks apply; crashed server sessions are reused on the next message
v2.1.251 (Aug 28)	Foreground subagent tool calls stream live to clients
v2.1.257 (Sep 1)	Dismissing the consent prompt no longer counts as consent
v2.1.273 (Sep 15)	Fork a session from the Claude app into a background session on your computer
v2.1.281 (Sep 23)	An unloaded org policy reads “couldn’t verify”, not “disabled”
v2.1.283 (Sep 25)	DISABLE_TELEMETRY / DO_NOT_TRACK no longer block it unless Trusted Devices is required
v2.1.284 (Sep 28)	In an untrusted folder, claude remote-control asks for workspace trust instead of exiting

Which remote surface for which job

Surface	Trigger	Claude runs on	Constraint	Pick it when ^[inferred]
Remote Control	Drive a running session from claude.ai/code or the app	Your machine (CLI, Desktop, VS Code)	claude process must keep running	Mid-task locally, want that context on your phone
Dispatch	Message a task from the Claude mobile app	Your machine (Desktop)	Computer awake with the Claude app open; Pro or Max	The job lives in desktop apps, not a code session
Channels	Events pushed from Telegram, Discord, iMessage or your own server	Your machine (CLI)	Events arrive only while the session is open; research preview (April 2026 docs)	CI failures or chats should reach a live session
Routines	Schedule, API call or GitHub event	Claude Code’s web infrastructure	Daily caps at launch: Pro 5, Max 15, Team/Enterprise 25	Recurring work while the laptop is closed
Claude Code on the web	Start a cloud session at claude.ai/code	A fresh cloud sandbox cloned from GitHub	Runs off your machine, Anthropic-managed by default	No local setup, uncloned repo, or parallel tasks

Practitioner patterns

• Sid Benesaria (a founding engineer of Claude Code) calls it his “favorite feature”: when Claude needs input, “your phone will buzz”.
• Chris turned on auto-start on a tip from Boris, Claude Code’s creator. He prefers it to Cursor’s cloud agents because “Claude Code has access to all of the MCPs and CLIs I have on my device”.
• A Code with Claude Tokyo presenter starts a long task, goes for a walk, and asks subagents from the phone to “spin up different tasks”.
• Anthropic’s London presenter recommends setting it “to always be on”.
• Nate’s non-coders course saw /clear sent from the phone arrive as “no content”, which is the bug fixed in v2.1.224.

Try It

1. Run claude update, then /login through claude.ai. Unset ANTHROPIC_API_KEY and ANTHROPIC_AUTH_TOKEN.
2. In a project, start claude, run /remote-control api-refactor, and select Enable Remote Control.
3. On your phone, open the Claude app, tap Code, and open the session with the green dot.
4. Turn on both push toggles in /config, then test by asking “notify me when the tests finish”.
5. For parallel work in a git repo, run claude remote-control --spawn worktree and start sessions from the app.

Implementation

Tool/Service: Claude Code (CLI, Desktop, VS Code) plus the Claude app or claude.ai/code. Setup: A claude.ai login on Pro, Max, Team or Enterprise. On Team and Enterprise, an Owner enables it at claude.ai/admin-settings/claude-code. Cost: The sources give no separate price; it requires one of these subscriptions. Integration notes: attribution.sessionUrl (v2.1.183) leaves the session link out of commits and PRs. claude doctor names the failed eligibility check.

claude remote-control --name "My Project" --spawn worktree   # server mode
claude --remote-control "My Project"                          # interactive (alias --rc)
claude remote-control --continue                              # bring back the server's first session

Open Questions

• Launch version and date: not in the vault’s changelogs. Their earliest Remote Control entry is v2.1.101 (Apr 10, 2026, “Multiple fixes for Remote Control”).
• Team and Enterprise: when did they become eligible?
• Other accounts: does a session URL refuse a different claude.ai account?
• Device cards: which release in v2.1.234–239 shipped them?

Related

• Claude Code Channels: pushes events into a local session.
• Dispatch in Claude Cowork: phone task, desktop execution.
• Claude Code Routines: cloud runs, no machine needed.
• Claude Code CLI Reference: all flags.
• Stop Babysitting Your Agents: Sid Benesaria’s talk.
• Chris’s AI Coding Workflow Update: auto-start and Cursor comparison.
• Pocket Access to Always-On Agents: SSH vs companion app vs chat channel.
• What’s New — Week 34: device cards, end of preview.