Claude PC Automation: The Complete 5-Feature Guide for 2026

Claude PC Automation: Dispatch, Computer Use, Loop & Channels Explained (2026 Guide)

In just two months, Claude has shifted from a coding assistant to a full-blown autonomous agent platform. Dispatch, Remote Control, Computer Use, Loop, and Channels — five distinct features that let Claude drive your machine on your behalf — have landed back-to-back between February and April 2026.

📑Table of Contents

1. Claude PC Automation: Dispatch, Computer Use, Loop & Channels Explained (2026 Guide)
2. Claude autonomous agent features: a complete 2026 map
3. Dispatch — fire off async tasks from your phone
4. Remote Control — real-time steering from web or mobile
5. Computer Use — Claude clicks, types, and drives your screen
6. Loop (/loop) — the recurring-task scheduler for Claude autonomous agents
7. Channels — control Claude from Telegram, Discord, or iMessage
8. Plan comparison, OS support, and use-case guide (2026)
9. Version requirements — Claude Code release timeline
10. Frequently asked questions
11. Final take — what Claude PC automation changed for me

I subscribe to Claude Max 20x and use Cowork daily. Since Loop started letting me schedule recurring tasks, my everyday routines have genuinely changed: morning PR reviews, overnight test runs, and periodic health checks now happen without me touching the keyboard. “Claude PC automation” is no longer science fiction — it’s a practical option you can steer from your phone.

The flip side is that there are now too many features to track. This guide organizes all five autonomous features, their differences, their plan and version requirements, and the situations each one is best suited for, so you can pick the right fit for your workflow.

Who this is for: Claude Pro / Max users and developers or business professionals evaluating AI-driven workflow automation.

Five-feature summary at a glance

Feature	What it does	Plan	OS	Status
Dispatch	Send async tasks from phone to PC	Pro / Max	macOS + iOS/Android	Research Preview
Remote Control	Drive Code session from web/mobile	Pro / Max	macOS	Research Preview
Computer Use	Claude clicks, types, controls apps	Pro / Max (no Team/Enterprise)	macOS only	Research Preview
Loop (/loop)	Schedule recurring tasks	All paid plans	All OS	GA
Channels	Control Claude from Telegram/Discord/iMessage	Pro / Max (admin toggle for Team/Ent)	All OS	Research Preview

Source: Claude Support, Anthropic Docs (as of April 2026).

Two months, five releases: a timeline

Claude’s autonomous feature set arrived in a remarkably tight window. Describing it as “a platform that grew up in two months” isn’t hyperbole.

---

Claude autonomous agent features: a complete 2026 map

The five features all “work on your behalf,” but they differ in how you trigger them and where execution actually happens. The mental model that worked for me is to split them into two camps: mobile-triggered agents (Dispatch, Remote Control, Channels) and desktop-side autonomous execution (Computer Use, Loop).

Canonical references live at Anthropic Docs and Claude Support. In my experience the gentlest onboarding path is to start with Loop first, then expand into Dispatch or Computer Use once you have a feel for the autonomy model.

---

Dispatch — fire off async tasks from your phone

🎯 Best for: developers who want to kick off builds, tests, or PRs away from the desk

Dispatch is the “tell Claude and walk away” feature. From a café or a train, you can send a prompt to Claude Code running on your Mac, it executes in the background, and a push notification tells you when the job is done.

Setup is literally one QR-code pairing between the Claude mobile app and your Mac. Anthropic launched it on March 17, 2026 as a Research Preview — initially Max-only, but expanded to Pro four days later.

The change in my workflow has been bigger than I expected. Tasks I used to batch until the evening — running the test suite, kicking off a long build, drafting PRs — now happen while I’m in transit, and I walk back to a machine that’s already ahead of me.

If you get stuck on the setup, our Claude Dispatch setup guide covers the full walkthrough and common error fixes.

---

Remote Control — real-time steering from web or mobile

🎯 Best for: continuing an in-progress session from elsewhere

Remote Control is the synchronous cousin of Dispatch. You start a claude session in your terminal, then hand it off to a web browser or mobile app and continue the conversation in real time.

The rule I use to decide between the two is simple: if I want to watch progress and steer, Remote Control; if I want to hand off and ignore, Dispatch. It shipped in February 2026 as a Research Preview for Pro and Max plans.

---

Computer Use — Claude clicks, types, and drives your screen

🎯 Best for: full automation of GUI-only apps (native IDEs, simulators, legacy tools)

Computer Use is the most autonomous feature of the five. The CLI version requires Claude Code v2.1.85 or later, runs on macOS only, and works exclusively in interactive sessions (the -p non-interactive flag is not supported).

Crucially, Claude doesn’t just start mashing the screen. Internally it works through a tool priority list and treats direct screen control as a last resort, reaching for it only when no cleaner tool is available.

Tool selection priority

Priority	Tool	Example
1 (highest)	MCP Connector	Direct APIs (Gmail, Drive, Slack)
2	Bash	Anything a CLI can finish
3	Claude in Chrome	Web browser tasks
4 (last resort)	Computer Use	GUI-only apps, native IDEs, simulators

Source: Claude Code Docs – Computer Use (as of April 2026).

macOS permissions you must grant

Two macOS permissions need to be granted to Claude Code before Computer Use can work.

1. Accessibility (for clicks, keystrokes, scrolling) — System Settings → Privacy & Security → Accessibility → enable Claude Code
2. Screen Recording (so Claude can see the screen) — System Settings → Privacy & Security → Screen Recording → enable Claude Code
3. Restart Claude Code — Screen Recording only takes effect after a restart
4. Enable the computer-use MCP server from the /mcp menu (off by default, enabled per project)

Per-app permission tiers

During a session, each app you allow Claude to touch lands in one of several permission tiers. These are applied automatically based on what the app is.

Tier	Apps	Capability
View-only	Browsers, trading platforms	Observe only; no clicks
Click-only	Terminal, iTerm, VS Code, Warp	Click allowed, flagged as shell-access equivalent
Full (flagged)	Finder (file I/O warning), System Settings (system-change warning)	Full control with warnings on sensitive actions
Full control	Everything else	Unrestricted after approval

Source: Claude Code Docs – Computer Use (as of April 2026).

Stop controls and safety design

Anthropic built in several ways to stop Claude the moment anything looks off.

• Esc key — globally consumed, so it works regardless of which app is focused (prompt-injection safe)
• Ctrl+C — stop from the terminal
• A persistent macOS notification shows “Claude is using your computer · press Esc to stop” throughout the session
• Screenshots are auto-downscaled (16″ MacBook Pro Retina 3456×2234 → ~1372×887) and the terminal itself is excluded from screenshots to prevent feedback loops
• Machine-wide session lock — only one Computer Use session can run at a time

Official example workflows

The Anthropic docs showcase three scenarios where Computer Use shines — all of them tasks a pure API couldn’t touch.

• Validate a native build — compile a Swift app, launch it, and have Claude visually confirm the UI
• Reproduce a layout bug — walk through a web or app UI bug manually and capture the state with screenshots
• Test a simulator flow — drive an iOS simulator for regression testing of screen transitions

---

Loop (/loop) — the recurring-task scheduler for Claude autonomous agents

🎯 Best for: anyone on a paid plan who wants a low-risk entry point to automation

/loop runs cron-style recurring tasks inside a Claude Code session. It has been GA since Claude Code v2.1.72 and is the most accessible autonomous feature — available on every paid plan.

Loop is the feature that most visibly changed my day-to-day. Since I started scheduling recurring jobs — morning PR reviews, nightly test runs, periodic monitoring checks — my workday starts with Claude already a few steps ahead. “Daily routines running on autopilot” sounds small until you actually live it.

# Run tests every 30 minutes
/loop 30m "Run tests and report failures"

# Review open PRs hourly
/loop 1h "Review open PRs"

# Run once every morning at 9 AM
/loop 09:00 "Summarize overnight activity"

Limits and the upgrades that replace them

• Session-bound — closing Claude Code resets your loops
• Up to 50 tasks, 3 days maximum — not designed for long-lived schedules
• Upgrade: Desktop schedule tasks — survive restarts, managed from Claude Desktop
• Upgrade: Cloud schedule tasks — no local machine needed; run in the cloud

My workflow is to experiment with /loop, promote proven jobs to Desktop schedule tasks, and migrate anything that shouldn’t depend on my local machine up to Cloud schedule tasks.

---

Channels — control Claude from Telegram, Discord, or iMessage

🎯 Best for: people who’d rather speak to Claude through their existing messaging apps

Channels lets you drive a Claude Code session from Telegram, Discord, or iMessage. It’s implemented as an MCP server, uses an allowlist to control who can send messages, and requires Claude Code v2.1.80 or later, the Bun runtime, and a claude.ai login.

PCMag framed Channels as Anthropic’s fast response to OpenClaw and Manus momentum. The integration is bidirectional — Claude’s replies come back in the same messaging channel you sent from.

“OpenClaw killer”? My verdict after using both

Channels has picked up the “OpenClaw killer” nickname in some corners of the community. Having used both in production, my answer is more nuanced: Channels partially replaces OpenClaw, but not entirely.

On raw capability, OpenClaw still wins — at the cost of security. Its browser flexibility, third-party integration breadth, and custom-workflow freedom can outreach Claude in situations where you’re willing to accept fewer guardrails.

On stability and day-to-day ergonomics, Claude is clearly ahead. As an officially maintained MCP server, authentication and messaging integration “just work.” For the things I rely on every day, Claude is the tool I trust.

Cost tips the scale further. OpenClaw-style tools generally bill against API usage, which adds up fast for heavy users. Claude bundles Dispatch, Computer Use, Loop, and Channels into a flat Pro / Max subscription — the harder you push it, the stronger the economic case.

---

Plan comparison, OS support, and use-case guide (2026)

Feature availability by plan

Feature	Free	Pro ($20)	Max 5x ($100)	Max 20x ($200)	Team	Enterprise
Dispatch	❌	✅	✅	✅	✅	✅
Remote Control	❌	✅	✅	✅	✅	✅
Computer Use	❌	✅	✅	✅	❌	❌
Loop (/loop)	❌	✅	✅	✅	✅	✅
Channels	❌	✅	✅	✅	△ admin toggle	△ admin toggle

Source: Claude Support (as of April 2026).

Which feature fits which job

Goal	Best feature
Run builds, tests, or PRs from your phone	Dispatch
Continue an in-progress session remotely	Remote Control
Automate GUI-only app flows	Computer Use
Morning PR reviews on autopilot	Loop / schedule tasks
Talk to Claude from Telegram/Discord	Channels

Price-to-value — which plan should you pick?

I’m on Claude Max 20x ($200/month) and I push it hard. Running Cowork, Dispatch, and Loop every day, plus Computer Use and Channels on top, Max 20x is where I stop bumping into “feels throttled” moments. It’s honest to say I use up most of the quota every week.

A simple rule of thumb for picking a plan:

• Trying things out / Loop-first: Pro ($20) is enough. Every feature is unlocked
• Dispatch + Computer Use used daily: Max 5x ($100) gives more breathing room
• Heavy everyday usage (my profile): Max 20x ($200) is the right call

Reaching the same capability via OpenClaw, Manus, or raw API agents generally means per-usage billing, which gets expensive fast. Claude bundles Dispatch, Computer Use, Loop, and Channels into the flat subscription, so the heavier your usage, the better the economics look. For a deeper plan breakdown, see our Is Claude Max worth it? analysis.

---

Version requirements — Claude Code release timeline

Each autonomous feature arrived in a specific Claude Code version. Here’s the compact reference.

Claude Code version	Feature added	Notes
v2.1.72+	Loop (/loop)	GA. Session-level cron
v2.1.80+	Channels	Bun runtime + claude.ai login required
v2.1.85+	Computer Use (CLI)	Interactive sessions only; -p non-interactive flag unsupported

Source: Claude Code Docs, Claude Support (as of April 2026).

---

Frequently asked questions

Q1: Which plan do I need to access every feature?

Pro ($20/month) unlocks Dispatch, Remote Control, Computer Use, Loop, and Channels. Max plans don’t unlock additional features — they raise usage limits. Heavy users like me still benefit from Max 5x or Max 20x, but functionality is identical on Pro.

Q2: Does any of this work on Windows?

The desktop side of Dispatch and the CLI version of Computer Use are macOS only. Loop and Channels work on all operating systems. Anthropic has announced Windows support as “coming soon” but without a firm date. Note: the API-level Computer Use (announced as a public beta on October 22, 2024) is OS-agnostic via the Anthropic API, but that’s a separate product surface from the Claude Code CLI integration covered in this article.

Q3: How is Dispatch different from Remote Control?

Dispatch is asynchronous — fire a task, walk away, wait for a notification. Remote Control is synchronous — real-time hand-off of an existing Code session that you actively steer. Pick Dispatch to run things in the background, Remote Control to stay in the loop while they run.

Q4: Why can’t I use Computer Use with the -p non-interactive flag?

Computer Use is explicitly designed as an interactive-session feature. The safety model assumes a human is watching, so Anthropic disabled it in non-interactive mode. If you want to bake screen automation into a CI pipeline, consider Playwright or similar tools instead.

Q5: How do I stop my Mac from sleeping while Dispatch or Computer Use runs?

Go to System Settings → Battery (or Energy Saver on desktops) and disable automatic sleep. For a one-off session, caffeinate -di is the standard trick. If you rely on these features daily, a dedicated always-on Mac mini is a common setup.

Q6: I granted Screen Recording but Computer Use still won’t see the screen

Screen Recording permission only takes effect after you restart Claude Code. Accessibility permission applies immediately, which makes the restart requirement for Screen Recording easy to miss.

Q7: Is Computer Use safe to use on financial or medical data?

Anthropic’s own guidance says no. Computer Use runs on the real, unsandboxed desktop. You can stop it with Esc, and Claude asks permission before touching each app, but for sensitive workflows you should validate it on a test account first. Treat Research Preview status seriously.

Q8: What happens if I try to run more than one session at once?

Computer Use uses a machine-level session lock — only one session can run per machine. That means you can’t run Dispatch and Computer Use CLI simultaneously. For shared Mac minis, coordinate with your team.

Q9: What’s the difference between /loop and schedule tasks?

/loop is session-scoped, capped at 50 tasks and 3 days. Desktop schedule tasks survive restarts; Cloud schedule tasks don’t even require your local machine. I use /loop to prototype, then promote proven jobs to Desktop or Cloud schedule tasks.

Q10: Is Channels really an OpenClaw replacement?

Partially. OpenClaw still covers more capability surface, but Claude wins on stability and flat-rate cost. I’ve been gradually shifting my main workflows over from OpenClaw to Claude Channels.

Q11: Any features locked out of Team or Enterprise?

Computer Use is currently unavailable on Team and Enterprise. Channels requires an admin toggle on those plans. On top of that, Team and Enterprise base plans don’t include Claude Code itself — you need to purchase premium seats for any user who will run Dispatch, Remote Control, Loop, or Channels (all of which route through Claude Code). Compare Pro vs. Team + premium seats carefully before rolling out to a team.

Q12: Where did you go when you first got stuck?

Honestly, I just went straight to the official docs and community usage examples — docs.anthropic.com and support.claude.com. They answer most first-week questions faster than any third-party tutorial. For Dispatch-specific setup headaches, our Dispatch setup guide walks through the flow step by step.

---

Final take — what Claude PC automation changed for me

• Dispatch — async task delivery from your phone
• Remote Control — real-time remote driving of a Code session
• Computer Use — full GUI automation on macOS (v2.1.85+)
• Loop — the low-risk entry point on every paid plan
• Channels — messaging-app control surfaces

Loop is running my daily routines, Dispatch is keeping things moving when I’m out, and Channels is gradually replacing my OpenClaw workflows. All of it is still a Research Preview, so expect fast change — but “doing complex work with a phone in one hand” has already stopped being a future tense phrase.

Related reading: Claude Dispatch setup guide, Claude pricing plan comparison, Is Claude Max worth it?, Claude Cowork workflow automation, and AI agents: complete guide.