Claude Dispatch: The Complete Guide — Control AI Agents From Your Phone (2026)

The LinkedIn Post Everyone Is Sharing Is Half Right

You've seen the post. It says "Claude Dispatch makes OpenClaw look like a relic." It shows people controlling AI agents from their phones, scraping hundreds of leads while waiting for the train.

The core idea is real. Anthropic genuinely shipped the ability to run and manage AI agents from your phone in March 2026.

But the post conflates three different things, names them incorrectly, and misses the most important part: there isn't one feature called "Claude Dispatch" — there are three distinct features, built for completely different users, and only one of them is actually called Dispatch.

Getting this distinction right is the difference between spending four hours trying to set up the wrong thing and having your actual workflow running in 45 minutes.

This guide covers all three. You'll know exactly which one you need, how to set it up correctly, how to secure it, and what OpenClaw can still do that Anthropic's tools cannot.

Why did Anthropic accelerate this work? OpenClaw. In early March 2026, nearly 1,000 people queued outside Tencent's Shenzhen headquarters to have OpenClaw installed on their laptops. Tencent announced an enterprise suite built on top of it, compatible with WeChat. That was the catalyst. Anthropic saw the demand for messaging-app AI control and shipped their version.

The Three Features — Which One Do You Actually Need?

Quick decision rule: Not a developer? → Use Dispatch. Developer who wants to watch/steer sessions? → Use Remote Control. Developer who wants Telegram/Discord integration during tasks? → Use Channels.

Part 1: Cowork Dispatch — Control Your Desktop From Your Phone

What It Actually Does

Here's the mental model that matters:

Your phone is a command surface. Your Mac is the execution engine. When you send a task from your phone, Claude on your desktop does the work using keyboard and mouse control — and sends you back the finished output. You don't watch every step. You come back to results.

Real tasks people run:

• "Organize the Downloads folder and move files older than 30 days to Archive"
• "Scan my last 50 emails, summarize anything needing action, draft replies"
• "Take the spreadsheet in ~/Documents/Q1-data.xlsx and build a summary table"
• "Check three URLs I'll paste and compare their pricing sections"
• "Pull the last week of server logs, flag errors, write a summary doc"

What Dispatch Uses Under the Hood

Dispatch is part of Claude Cowork — Anthropic's desktop app aimed at knowledge workers who aren't developers. It uses computer use (Claude controlling your keyboard and mouse) when needed, plus file system access, calendar, email, and any connectors you configure.

Setup: Cowork Dispatch

Requirements:

• macOS (Windows not supported at launch)
• Claude Pro ($20/mo) or Max ($100/mo) subscription
• Claude Desktop app installed
• Claude mobile app on iOS or Android

Step 1: Install the Claude Desktop App

Download from claude.ai/download. During first launch, you'll be asked to sign in and grant permissions.

Step 2: Open Cowork and Enable Dispatch

1. Open Claude Desktop
2. Navigate to Cowork in the sidebar
3. Find Dispatch and click "Get started"
4. Grant the requested file access permissions
5. Enable the option to keep your computer awake while Dispatch is active

Step 3: Pair Your Phone

A QR code will appear on your desktop. Open the Claude mobile app on your phone and scan it. Your phone and desktop are now linked to a single persistent Dispatch thread.

Step 4: Test With a Simple Task

Open the Claude app on your phone, go to the Dispatch thread, and type:
"Create a file called dispatch-test.txt on my Desktop with today's date and the message 'Dispatch is working'"

If a file appears on your Desktop, your setup is complete.

Security: The Computer Use Risk

Practical security rules for Dispatch:

1. Grant access only to specific folders — not your entire home directory
2. Don't point it at folders containing .env files, SSH keys, password manager databases, or API keys
3. Don't use it to process documents you received from unknown sources
4. Disable Dispatch when not actively using it
5. Review the access permissions list regularly

Part 2: Claude Code Remote Control — Steer Coding Sessions From Anywhere

This is the feature most developers actually want. If you run Claude Code and have ever needed to check in on a long-running task while away from your desk, Remote Control is for you.

What It Actually Does

┌──────────────────────────────────────────────────┐
│  Your Machine (terminal running Claude Code)     │
│                                                  │
│  $ claude --remote-control "my-project"          │
│  > Starting remote session...                    │
│  > QR code: [scan with phone]                    │
│  > Session URL: https://claude.ai/code/r/abc123  │
│                                                  │
│  [Claude Code continues running normally]        │
└──────────────────────────────────────────────────┘
         ↓ outbound HTTPS only (no open ports)
┌──────────────────────────────────────────────────┐
│  Your Phone / Any Browser                        │
│                                                  │
│  → See real-time output                          │
│  → Send new instructions                         │
│  → Approve permission requests                   │
│  → Navigate conversation history                 │
└──────────────────────────────────────────────────┘

Your machine never opens inbound ports. Claude Code makes outbound HTTPS connections to Anthropic's API as a relay layer. Your browser or phone connects to the session through that relay. If the relay drops for more than 10 minutes, the session terminates.

What you can do from your phone:

• Watch Claude Code work in real time
• Send new messages and instructions mid-task
• Approve tool use requests that need your permission
• Review what Claude has written or changed
• See errors as they happen

What you cannot do from your phone:

• Run this without a paid claude.ai account (API keys don't work — requires OAuth)
• Use this if you're on Bedrock, Vertex, or Azure Foundry (not supported)
• Keep the session alive if the terminal closes or your machine sleeps

Setup: Remote Control

Requirements:

• Any paid Claude plan (Pro, Max, Team, or Enterprise)
• Claude Code v2.1.51 or later
• macOS, Linux, or Windows (WSL)
• claude.ai OAuth login (not API key)

Check your version:

claude --version

# Update: npm update -g @anthropic-ai/claude-code

Method 1: Start a fresh remote-enabled session

claude --remote-control "my-project-name"

Claude Code starts and immediately shows a QR code and a URL. Scan the QR code with your phone or open the URL in any browser.

Method 2: Enable remote control on an existing session

# Inside a running Claude Code session:
/remote-control my-project-name
# Press spacebar to toggle the QR code display

Method 3: Dedicated server mode (for long-running tasks)

claude remote-control --name "background-task"
# Starts in server mode — no initial conversation, waits for remote input

Confirming it works:

1. Run one of the commands above
2. Open the URL in your phone browser or Claude mobile app
3. You should see the Claude Code conversation interface
4. Send "hello" from your phone — Claude Code on your machine should respond

Security: The Architecture Is Safe by Design

Remote Control was designed with a specific security property: your machine never becomes a server.

• No inbound ports are opened on your machine
• Connections flow: Machine → Anthropic relay → Browser/Phone
• Each session uses short-lived, single-purpose credentials
• Sessions have a hard 10-minute network timeout
• Requires claude.ai OAuth — cannot be accessed with a stolen API key alone

The main risk isn't the architecture — it's what you leave running. If you start a Claude Code session with --dangerously-skip-permissions and then enable Remote Control, anyone who gets the session URL has unsupervised access to your terminal.

Security practices:

# Never combine these two flags:
claude --dangerously-skip-permissions --remote-control "session"
# ↑ This gives remote access with no permission gates. Don't do this.

# Safe pattern: keep permission gates on, approve from phone
claude --remote-control "session"
# Approve each permission request from your phone as it comes in

Part 3: Claude Code Channels — Telegram and Discord Inside Your Coding Sessions

Channels is the most experimental of the three features and the one that directly answers the "OpenClaw killed" framing. It lets Claude Code communicate with you through messaging apps while it works — and lets you respond back.

What It Actually Does

┌────────────────────────────────────────────────┐
│  Claude Code (running on your machine)         │
│  [executing a long task — code refactor, etc.] │
└───────────────────┬────────────────────────────┘
                    ↓
          Plugin: telegram@claude-plugins-official
                    ↓
┌────────────────────────────────────────────────┐
│  Your Telegram Bot                             │
│                                                │
│  Claude: "Finished refactoring auth module.    │
│  Found 3 files with deprecated patterns.       │
│  Want me to update those too?"                 │
│                                                │
│  You: "yes, go ahead"                          │
│                                                │
│  Claude: "Done. Running tests now..."          │
└────────────────────────────────────────────────┘

This is bidirectional. Claude Code can send you updates at milestones and ask for decisions. You can send back instructions or approvals. It works over Telegram and Discord — those are the only two platforms supported in the March 2026 research preview.

Setup: Channels (Telegram)

Requirements:

• Claude Pro or Max plan
• Claude Code v2.1.80 or later
• claude.ai OAuth login (API keys not supported)
• A Telegram account

Step 1: Update Claude Code and verify version

npm update -g @anthropic-ai/claude-code
claude --version
# Need v2.1.80 or higher

Step 2: Create a Telegram bot

1. Open Telegram and search for @BotFather
2. Send /newbot
3. Follow the prompts: give it a name and a username
4. BotFather returns a bot token — looks like 7143921044:AAFxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
5. Copy this token — you'll need it in the next step

Step 3: Install and configure the plugin

Start Claude Code (regular session):

claude

Inside the session, run these commands one at a time:

/plugin install telegram@claude-plugins-official

/reload-plugins

/telegram:configure 7143921044:AAFxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Replace the token with your actual BotFather token.

Step 4: Restart with Channels enabled

Exit the current session (Ctrl+C) and restart with the channels flag:

claude --channels plugin:telegram@claude-plugins-official

Step 5: Pair your account

1. Open Telegram and find your bot (by its username)
2. Send any message to start a conversation with it
3. Back in Claude Code, run:

/telegram:access pair <code>

Replace <code> with the message Claude Code gives you when you ask it to start pairing.

Step 6: Lock it to your account only (critical security step)

/telegram:access policy allowlist

This makes your bot silently ignore every message from any account except your own. Do this before you do anything else. An unlocked bot is a risk.

Verify it works:

From Telegram, message your bot: ping

Claude Code should reply: pong

Setting Up Discord Instead

The process is nearly identical but uses the Discord plugin:

# Inside Claude Code:
/plugin install discord@claude-plugins-official
/reload-plugins
/discord:configure YOUR_DISCORD_BOT_TOKEN

# Restart:
claude --channels plugin:discord@claude-plugins-official

# Lock access:
/discord:access policy allowlist

Create a Discord bot at discord.com/developers/applications. Under the Bot section, get the token and enable "Message Content Intent."

How to Use Channels in Practice

Once running, Channels changes how you interact with long Claude Code tasks:

Milestone updates:

# Tell Claude to send updates at key points:
"Work through the test suite. Send me a Telegram message when you finish
each major section, and if you hit an error you can't resolve, ask me."

Remote approval gates:

# Set up a task where Claude pauses for permission:
"Refactor the API client module. Before running git commit, message me
on Telegram with a summary and wait for my 'go ahead'."

Background tasks with check-ins:

# Start a background session with Channels:
claude --channels plugin:telegram@claude-plugins-official

# From inside:
"Run the full migration script. Every 10 minutes, send me a status update
on Telegram. If anything fails, stop and tell me immediately."

The Honest Comparison: Anthropic vs. OpenClaw

What Anthropic Got Right

Setup simplicity. Remote Control is genuinely one command. Channels takes 30 minutes. Dispatch is a QR code scan. OpenClaw at full configuration requires a VPS, SSH tunneling, reverse proxy setup, Telegram bot configuration, and hours of troubleshooting. The difference in time-to-working is massive.

Security infrastructure. Anthropic provides SOC 2 Type 2 and ISO 27001 compliance. OpenClaw has 288 documented security notices on its GitHub repository and was flagged by Cisco and BitSight as a risk for casual users. The ClawHavoc incident (354 malicious skills uploaded to ClawHub in January 2026) infected users with Atomic Stealer, keyloggers, and RATs.

The developer workflow. Remote Control is genuinely excellent for what it does. Being able to steer a running Claude Code session from your phone — see real-time output, approve permissions, send new instructions — is a qualitative improvement for any developer who runs long-running tasks.

What OpenClaw Still Does Better

Platform reach. Channels supports Telegram and Discord. OpenClaw supports 20+ platforms: WhatsApp, iMessage, Signal, Slack, Teams, LINE, and more. If your team uses Slack or your family uses WhatsApp, Channels doesn't solve that problem.

True always-on operation. OpenClaw runs as a persistent background daemon. It doesn't need you to keep a terminal open. It doesn't need your laptop to stay awake. It checks in on a cron schedule whether you're at your desk or not. Anthropic's tools are fundamentally session-based — you initiate the session, you end it.

Scheduled proactive tasks. OpenClaw's HEARTBEAT.md runs tasks on a schedule without you being present. "Every morning at 7am, check my calendar and email and send me a priority summary." This is the original OpenClaw use case that went viral — and it's the one thing Anthropic's current tools genuinely don't replicate.

Multi-model and free. OpenClaw costs nothing and runs GPT-4o, Gemini, DeepSeek, and local models through Ollama. Anthropic's tools require a $20/month minimum and only use Claude.

The honest conclusion: Anthropic matched OpenClaw's headline feature (phone control), made it dramatically easier to set up, and made it significantly more secure. For most developers with a Claude subscription, Anthropic's tools are now the right choice for the specific use cases they cover. OpenClaw retains meaningful advantages for power users who need always-on operation, broad messaging platform support, multi-model flexibility, or zero subscription cost.

Real-World Workflows: What to Actually Build

Workflow 1: The Long Build Monitor (Remote Control)

You've kicked off a large refactor or a slow test suite. You need to step away from your desk.

# Start Claude Code with remote control
claude --remote-control "build-monitor"

# Inside the session, set up the task:
"Run the full test suite. Give me updates every time a major section passes
or fails. If anything fails three times in a row, stop and tell me exactly
what happened. I'll be checking in from my phone."

Open the session URL on your phone. Check in every 20 minutes. Send corrections or approvals as needed. Come back to your desk when it's done.

Workflow 2: The Document Assistant (Dispatch)

You're traveling and need work done on your Mac back home.

From your phone, in the Dispatch thread:
"There's a folder on my Desktop called Q1-Reports. Open each file, extract
the key metrics, and create a summary document called Q1-Summary.md with
a table. Email it to me when done."

Your Mac handles it. You check back when you're back at the hotel.

Workflow 3: The Background Code Task With Updates (Channels)

You're running a migration script that takes two hours and needs one decision point.

# Start Claude Code with Telegram channel
claude --channels plugin:telegram@claude-plugins-official

# In Claude Code:
"Run the database migration script in /scripts/migrate.py.
After the first 500 rows are migrated, send me a Telegram message with
a sample of 5 rows so I can verify the format looks right.
Wait for my reply before continuing. When it's fully done, send me
a final count."

You get a Telegram message mid-task, verify the output looks right, reply "looks good, continue", and get a final summary when it finishes.

Workflow 4: The Parallel Research Session

Two Claude Code sessions running simultaneously, one for each research thread:

# Terminal 1:
claude --remote-control "research-session-a" &

# Terminal 2:
claude --remote-control "research-session-b" &

You check both sessions from your phone, send corrections to whichever needs attention, leave them running.

Troubleshooting: The Problems You'll Actually Hit

The Version Requirements, Summarized

# Check and update Claude Code:
claude --version
npm update -g @anthropic-ai/claude-code
claude --version   # confirm the new version

Where Each Feature Is Headed

Anthropic has been explicit that all three features are in early form. Based on the research preview documentation and community feedback threads, here's what's actively being worked on:

Dispatch: macOS-only is a temporary limitation. Windows support is on the roadmap. The ~50% success rate with computer use tasks is a known issue — it's directly tied to Claude's computer use capability improving. Expect both to improve through 2026.

Remote Control: This feature is already generally available and considered stable. The most-requested addition is persistent sessions that survive terminal restarts — currently sessions die with the terminal.

Channels: WhatsApp and Slack are the most-requested additions. Anthropic has not committed to a timeline. The research preview constraint (Anthropic-approved plugins only) will likely loosen as the feature matures.

The trajectory is clear: Anthropic is building toward a model where Claude Code is something you can start on your machine, leave running, and stay connected to from anywhere. These three features are the beginning of that architecture — not the end state.

What You Should Do Right Now

The idea in the LinkedIn post is real: you can run and manage AI agents straight from your phone. No OpenClaw required, no VPS to configure, no reverse proxy to wrestle with.

The catch is that what LinkedIn calls "Claude Dispatch" is actually three separate features, each solving a different problem. Remote Control is for developers who want to watch and steer coding sessions. Channels is for developers who want Telegram or Discord integration. Dispatch is for knowledge workers who want to delegate desktop tasks.

Know which one you need. Set up only that one. Secure it before you use it for anything real.

The tooling is genuinely good — and it keeps getting better.