Summary: Claude Code Remote Control arrived on February 24, 2026 and lets claude.ai or mobile apps supervise a local session. Code execution and filesystem access remain on the user’s machine, while permission prompts can be approved remotely. The article notes compliance limits: transcripts reside with Anthropic during connections and direct access to api.anthropic.com is required.
Remote Control for Claude Code arrived with little fanfare from Anthropic on February 24, 2026. Within a day, Simon Willison published a hands-on account: fire off claude remote-control in a terminal, and the Claude Code session living on your machine can then be driven from the claude.ai web app or the iOS and Android apps — the phone lists it as “Remote Control Session (Mac)”. If Claude Code is already where you spend your working day, this closes the one gap that remained: once you step away from the terminal, the phone keeps you watching the agent.
The architectural tradeoff deserves a second look. Nothing about code execution or filesystem access ever leaves your machine — the remote surfaces amount to a window. Managed offerings that relocate the work into the cloud are making the opposite wager.
The Mechanics: Local Execution, the Remote End a Window
According to the official documentation, Remote Control connects the claude.ai/code web interface or the Claude mobile apps to a Claude Code session running on your machine. Your local environment survives intact — MCP servers, custom tools, and project config behave exactly as before — while the conversation stays in sync across terminal, browser, and phone. Mobile can push image and file attachments into the session, and when the network drops it finds its way back on its own, delivering whatever messages queued up in the meantime.
Entry points come in three flavors. claude remote-control launches a dedicated server process and understands flags such as —name, —continue, and —capacity (defaulting to 32 sessions). claude —remote-control grafts remote control onto a session you are already interacting with. For a session already running, /remote-control switches it on without losing the conversation history. The VS Code extension takes the same commands through its prompt box.
The boundaries are spelled out just as plainly: an interactive process carries a single remote session (server mode is the answer beyond that), the session drops offline the instant its local process exits, and server mode gives up after roughly ten minutes without network. For SSH setups, the docs recommend keeping it alive under tmux or screen.
The Permission and Security Model
Security is where this release carries the most substance. Permission prompts and AskUserQuestion dialogs get relayed to whatever devices are connected and remain open until someone answers — so tool calls can be approved straight from a phone, with push notifications carrying pending approvals to the mobile side.
Networking-wise, the connection is strictly outbound HTTPS with zero inbound ports; traffic rides TLS through the Anthropic API behind short-lived, purpose-scoped credentials. Two caveats come attached: transcripts sit on Anthropic servers for as long as you are connected (governed by the data usage policy), which locks out organizations with Zero Data Retention requirements; and access has to reach api.anthropic.com directly — anyone routing through Amazon Bedrock, Google’s Agent Platform, Microsoft Foundry, or a custom gateway is out of luck for now. Plan coverage extends to Pro, Max, Team, and Enterprise; an API-key login does not qualify, and Team/Enterprise plans add an administrator toggle on top. The enterprise Trusted Devices beta stacks device enrollment onto a sign-in no older than 18 hours, refreshed via Face ID or a passkey.
Simon Willison Takes It for a Spin: Usable, Still Rough
Willison’s account doubles as a reminder of how early this release is. Attempt number one ran into “Remote Control is not enabled for your account” even though he administers the account himself — a logout and login in the terminal app cleared it up. The —dangerously-skip-permissions flag was accepted without complaint and did nothing, leaving every action to be approved by hand. Mid-run he landed in a state where every prompt came back with an API 500, and once he restarted the program, the old sessions answered with cryptic API errors rather than a plain statement that the session had ended. His verdict pulls no punches: “It’s a little bit janky right now,” though he reckons the fixes will come fast.
The OpenClaw Contest and Cowork’s Scheduling
Willison frames the feature against a wider competitive backdrop: steering the agent on your own device from a phone is exactly what OpenClaw-style tools sell as their core pitch, and Anthropic has now folded that same capability into the mainstream toolchain. He also names the gap Claude Code has yet to close — “the other killer feature of the Claw category of software”: scheduling.
Anthropic’s countermove came quickly. That same week brought the announcement of scheduled recurring tasks in Cowork, Claude Code’s general-agent sibling — with a constraint that bites: tasks run only while the computer is awake and the Claude Desktop app is open, and a missed task fires once when the machine wakes or the app comes back. Willison’s two cents: “I really hope they’re working on a Cowork Cloud product.”
What This Changes for Developer Workflows
Three things worth taking away. First, long-running agent workflows gain an official channel for mobile supervision — approval friction, the part of agentic coding that breaks flow most, shifts onto the phone, and with tmux or server mode a commute becomes task-advancing time. Second, security and compliance teams owe Claude Code a fresh evaluation: transcripts residing on Anthropic servers while connected, together with the direct-to-api.anthropic.com requirement, will move adoption decisions at some enterprises. Third, the local-versus-cloud question for agents gets sharper — Anthropic’s bet for now is local execution paired with remote control, and whether a cloud edition (the Cowork Cloud Willison is calling for) ever materializes is what to watch next.