On February 24, 2026, Anthropic quietly shipped Remote Control for Claude Code. Simon Willison had a hands-on writeup the next day: run claude remote-control in your terminal, and that local Claude Code session becomes drivable from the claude.ai web app or the iOS/Android apps — on the phone it shows up as “Remote Control Session (Mac)”. For developers who already live in Claude Code, this fills in the last missing piece: keep watching the agent from your phone after you walk away from the terminal.
The design tradeoff is worth pausing on. Code execution and filesystem access stay on your machine the entire time; the remote surfaces are just a window. That is the opposite bet from managed offerings that move the work into the cloud.
How It Works: Local Execution, the Remote Is Just a Window
Per the official docs, Remote Control links the claude.ai/code web interface or the Claude mobile apps to a Claude Code session running on your machine. You keep your full local environment — MCP servers, custom tools, project config all work as usual — with the conversation synced across terminal, browser, and phone. You can send image and file attachments from mobile, and after a network drop it reconnects automatically and delivers queued messages.
There are three ways in. claude remote-control starts a dedicated server process supporting flags like –name, –continue, and –capacity (default 32 sessions). claude –remote-control bolts remote control onto an interactive session. An existing session can turn it on with /remote-control, preserving conversation history. The VS Code extension accepts the same command in its prompt box.
The limits are equally explicit: one remote session per interactive process (server mode for more), the session goes offline the moment the local process dies, and server mode exits after roughly ten minutes without network. The docs suggest tmux or screen for SSH setups.
Permissions and the Security Model
The security design is the most substantive part of this release. Permission prompts and AskUserQuestion dialogs are forwarded to connected devices and stay open until answered — meaning you can approve tool calls from your phone, with push notifications delivering pending approvals to mobile.
On the network side it is outbound HTTPS only, no inbound ports; traffic travels over TLS through the Anthropic API using short-lived, purpose-scoped credentials. Two caveats: transcripts are stored on Anthropic servers while connected (under the data usage policy), so organizations with Zero Data Retention requirements cannot enable it; and access must hit api.anthropic.com directly — users on Amazon Bedrock, Google’s Agent Platform, Microsoft Foundry, or a custom gateway are currently out of luck. Plan support covers Pro, Max, Team, and Enterprise; API-key logins do not count, and Team/Enterprise add an admin toggle. The enterprise Trusted Devices beta layers on device enrollment plus a sign-in no older than 18 hours, refreshed with Face ID or a passkey.
Simon Willison’s Field Test: Usable, but Janky
Willison’s first-hand experience is a reminder that this is an early release. His first attempt hit “Remote Control is not enabled for your account” despite being his own administrator — logging out and back into the terminal app fixed it. The –dangerously-skip-permissions flag was silently accepted but had no effect, so every action needed manual approval. At one point every prompt returned an API 500, and after restarting the program the old sessions produced cryptic API errors instead of saying the session had ended. His verdict is blunt: “It’s a little bit janky right now,” though he expects quick fixes.
The OpenClaw Race and Cowork Scheduling
Willison places the feature in a larger competitive frame: controlling the agent on your personal device from your phone is precisely the core pitch of OpenClaw-style tools, and Anthropic has now built the same capability into the mainstream toolchain. He also names what Claude Code still lacks — “the other killer feature of the Claw category of software”: scheduling.
Anthropic’s answer arrived fast. The same week it announced scheduled recurring tasks in Cowork, Claude Code’s general-agent sibling, with a real constraint: tasks only run while your computer is awake and the Claude Desktop app is open; a missed task runs once when the machine wakes or the app reopens. Willison’s comment: “I really hope they’re working on a Cowork Cloud product.”
Practical Impact on Developer Workflows
Three takeaways. First, long-running agent workflows now have an official mobile supervision channel — approval friction, the most flow-breaking part of agentic coding, moves to your phone, and with tmux or server mode you can advance tasks from a commute. Second, security and compliance teams need to re-evaluate Claude Code: transcripts living on Anthropic servers while connected, plus the direct-to-api.anthropic.com requirement, will change adoption decisions at some enterprises. Third, the feature sharpens the local-versus-cloud split for agents — Anthropic’s current bet is local execution with remote control, and whether a cloud edition (the Cowork Cloud Willison is calling for) materializes is the next thing to watch.
Sources
- Claude Code Remote Control — Simon Willison’s Weblog
- Claude Code Remote Control — official docs
- Schedule recurring tasks in Cowork — Claude Support
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.