Anthropic just expanded what "remote work" means for AI. As of this week, Claude can physically operate your Mac — opening files, clicking through browsers, filling spreadsheets — while you send instructions from your phone. Paired with two companion features called Dispatch and Remote Control for Claude Code, Anthropic has quietly built the most complete phone-to-desktop AI agent pipeline available today. This is a research preview, not a finished product, and it comes with real limitations. But for Pro and Max subscribers on macOS, it changes what a solo developer or knowledge worker can accomplish without sitting at a desk.
What You Need to Know
There are actually three distinct "remote control" capabilities Anthropic has launched in the past month. They are often conflated online but serve different users:
- Claude Code Remote Control — connects a terminal session to your phone or any browser. For developers.
- Dispatch in Claude Cowork — lets you text tasks to your Mac from your phone. For knowledge workers.
- Computer Use — lets Claude physically control your Mac (mouse, keyboard, browser) when no connector is available. Powers both of the above.
Switch to these features if you regularly start long tasks at your desk and need to step away before they finish. Stay put if you are on Windows, using an API key only, or on a Team/Enterprise plan — none of these are supported yet.
What Launched and When: A Timeline
| Feature | Launch Date | Available In | Plans |
|---|---|---|---|
| Claude Code Remote Control | February 24–25, 2026 | Claude Code (terminal) | Pro, Max |
| Dispatch | ~March 15, 2026 | Claude Cowork | Pro, Max |
| Computer Use (Mac) | March 23–24, 2026 | Cowork + Claude Code | Pro, Max |
| Windows support | Not announced | — | — |
| Team / Enterprise Dispatch | Not announced | — | — |
All three are research previews. Anthropic is explicit that complex workflows may fail and is actively gathering feedback.
What's New: Breaking Down Each Feature
Claude Code Remote Control
Remote Control connects claude.ai/code or the Claude mobile app (iOS and Android) to a Claude Code session running on your machine. When you enable it, your local process establishes an outbound HTTPS connection to Anthropic's API. From that point on, any device with access to the session URL — or the Claude app — can view the conversation and send messages, which are relayed back to your local session.
The critical architecture detail: nothing moves to the cloud. Your local Claude Code session makes outbound HTTPS requests only. No inbound ports open on your machine. Files and MCP servers never leave your machine. Only chat messages and tool results flow through the encrypted bridge.
This matters because your local setup is irreplaceable. Your CLAUDE.md configuration, file access, MCP integrations, and custom tools all remain available — a cloud session starts fresh.
With Remote Control you can use your full local environment remotely — your filesystem, MCP servers, tools, and project configuration all stay available. The conversation stays in sync across all connected devices, so you can send messages from your terminal, browser, and phone interchangeably. If your laptop sleeps or your network drops, the session reconnects automatically when your machine comes back online.
How to Set It Up
Remote Control requires Claude Code v2.1.51 or later. You can start a dedicated Remote Control server, start an interactive session with Remote Control enabled, or connect a session that's already running. The process stays running in your terminal in server mode, waiting for remote connections. It displays a session URL you can use to connect from another device, and you can press spacebar to show a QR code for quick access from your phone.
To enable it mid-session, use the /remote-control or /rc slash command — it carries over your full conversation history.
Known Limitations
If your machine loses network connectivity for more than roughly 10 minutes, the session times out and the process exits. Short interruptions — like your laptop sleeping briefly or a momentary Wi-Fi drop — are handled gracefully with automatic reconnection. Multi-machine setups may show inconsistencies: sessions from some machines persist in the app's session list while others may not.
Push notifications are not built in at the operating system level as of March 2026, so you need to periodically check the session rather than relying on alerts.
Dispatch in Claude Cowork
Dispatch extends computer use from a desktop-only tool into a remote delegation system. Released the week prior to March 22, 2026, Dispatch creates one persistent conversation thread that runs across your phone and your Mac simultaneously. You assign a task from your phone, redirect your attention, and return to completed work on your desktop.
From your phone, you can hand Claude tasks that use everything on your desktop, including things you can't open on your phone. For example: ask Claude to pull data from a local spreadsheet and compile a summary report, have Claude search your Slack messages and email then draft a briefing document, or request a formatted presentation built from files in your Google Drive.
Think of Dispatch as the non-technical version of Claude Code Remote Control. Same idea — phone controls desktop — but oriented around document and productivity workflows rather than terminal sessions.
Computer Use: Claude Controls the Mouse and Keyboard
This is the newest and most dramatic addition. When Claude doesn't have an existing connector for a service — like Slack or Google Calendar — it can directly control your browser, mouse, keyboard, and screen to complete tasks. It scrolls, clicks to open, and explores as needed, always asking for your explicit permission first.
Anthropic's demo shows a user asking Claude to export a pitch deck as a PDF and attach it to a meeting invite, all while the user is away from their Mac.
Computer Use is the fallback layer. Claude always tries service connectors first. Only when those do not exist does it reach for direct screen control. This tiered approach reduces unnecessary autonomy while ensuring tasks still complete.
Feature Comparison Table
| Capability | Claude Code Remote Control | Dispatch (Cowork) | Computer Use |
|---|---|---|---|
| Primary audience | Developers | Knowledge workers | Both |
| Interface | Terminal → Phone/Browser | Phone → Desktop | Underlying layer |
| Execution location | Your local machine | Your local machine | Your local Mac |
| OS support | macOS (Windows: check docs) | macOS, Windows x64 | macOS only |
| Cloud migration | None | None | None |
| Requires desktop awake | Yes | Yes | Yes |
| Push notifications | No | No | No |
| Multi-session support | One session per machine | One thread only | N/A |
| Plan required | Pro or Max | Pro or Max | Pro or Max |
| API key support | No | No | No |
Plan and Pricing Breakdown
| Plan | Monthly Price | Remote Control | Dispatch | Computer Use | Notes |
|---|---|---|---|---|---|
| Free | $0 | ✗ | ✗ | ✗ | Chat only |
| Pro | $20 | ✓ | ✓ | ✓ | Research preview |
| Max 5x | $100 | ✓ | ✓ | ✓ | Higher usage limits |
| Max 20x | $200 | ✓ | ✓ | ✓ | Highest usage limits |
| Team | $20/seat | ✗ | Not yet | ✗ | Dispatch expected |
| Enterprise | Custom | ✗ | Not yet | ✗ | Dispatch expected |
Pricing as of March 2026. Agentic tasks consume more quota than standard chat — some Max subscribers have reported heavy usage depletion from Dispatch sessions alone.
The Surprising Finding: One Feature, Two Very Different Products
Most coverage treats "Claude remote control" as a single thing. It is not. Claude Code Remote Control and Dispatch are architecturally similar but serve fundamentally different audiences with different setups, and if you're already using Claude Code, you probably don't need Dispatch. But if you've been using Cowork for file organization, research, or document workflows, Dispatch extends that workflow to your pocket.
The distinction matters for deciding which feature to set up. Developers should go straight to claude remote-control in the terminal. Non-developers using Cowork should go to the Dispatch tab in the desktop app. Attempting to set up the wrong one will likely produce the kind of friction Simon Willison described: initially getting the error "Remote Control is not enabled for your account. Contact your administrator." — even when you are your own administrator. The fix is logging out and back in through the Claude Code terminal, but that is not obvious from the documentation.
Security Model: What You Should and Should Not Trust
The rise of Claude remote control capabilities has intensified discussion around agentic AI security and user protection. Anthropic says it has implemented multiple safeguards, including automated checks designed to detect prompt injection attempts that might redirect the system into unsafe behavior. Certain applications that handle especially sensitive data are disabled by default to reduce exposure.
Anthropic is explicit in its guidance: do not use computer use with financial accounts, legal documents, medical information, or apps with others' personal data.
The security architecture for Remote Control is well-designed at the transport level — outbound-only HTTPS, no open ports, TLS encryption, short-lived scoped credentials. The risk is not data interception. The risk is agentic scope: giving a mobile AI agent remote control of a desktop AI agent creates a chain where instructions from your phone can trigger real actions on your computer — including reading, moving, or deleting local files, interacting with connected services, and controlling your browser.
Treat these tools like giving a capable but fallible contractor the keys to your office. Useful for defined tasks. Not appropriate for access to your most sensitive systems.
Who Should Use This — and Who Should Wait
Use Claude Code Remote Control now if you are a developer who regularly starts long-running Claude Code sessions — refactors, test suites, build pipelines — and needs to step away before they finish. The QR code setup takes under two minutes, the session persists through brief disconnects, and the mobile interface gives you diff-view approval of every file change.
Use Dispatch now if you are a knowledge worker on Cowork who wants to delegate recurring tasks — weekly reports, email summaries, file organization — without staying at your desk. Simple tasks like "summarize my emails" work reliably. Complex multi-step chains are still prone to failure at this stage. Start simple.
Enable Computer Use cautiously if you regularly need to interact with tools that lack Claude connectors. Let Claude try connectors first — it does this by default. Only turn on Computer Use explicitly for workflows where you have verified the task scope is appropriate.
Wait if you are on Windows (Computer Use is macOS-only, though Dispatch works on Windows x64 via Cowork), on a Team or Enterprise plan (Dispatch access is not yet available), or using an API key without a claude.ai subscription.
Hold off entirely if you manage sensitive financial, legal, or medical data in your local environment. Anthropic's own guidance is clear here, and the research preview label means the safeguards are still being hardened.
What to Watch Next
Windows support for Computer Use is the obvious gap — Anthropic has confirmed it is coming but has not given a timeline. Team and Enterprise access to Dispatch is similarly unannounced but likely near-term given that Cowork already supports those tiers. Push notifications for Remote Control sessions are a frequently requested feature that would significantly improve the "assign and walk away" workflow. The combination of Computer Use and Remote Control — sending instructions via phone while Claude executes desktop operations on your Mac — is one of the most anticipated feature pairings heading into Q2 2026.
Conclusion
Anthropic has shipped three interconnected features — Remote Control, Dispatch, and Computer Use — that together let Claude act as a genuine background worker on your Mac while you direct it from your phone. The developer path (Claude Code Remote Control) and the productivity path (Dispatch) are separate tools for separate audiences, and understanding which one applies to you is the first step to getting value from either. Both are research previews with real rough edges: no push notifications, macOS-only for Computer Use, one session per machine, and a quota cost that adds up fast on heavy workflows. For Pro and Max subscribers with compatible setups, the core capability is real and ready to test — start with a simple, reversible task, verify it works, and build from there.
