How to Use Claude Code Dispatch and Remote Control

Summary:

1. You’re building one feature at a time. Power users build three in parallel. The gap is four features you’re not using yet.
2. Dispatch runs parallel Claude instances. Three features at 45 minutes sequential = 15 minutes dispatched.
3. Remote control lets you approve changes from your phone while Claude works on your machine.
4. Adoption timeline, planning method, and comparison grid so you know what to use this week and what to ignore until next month.

You finish a feature. Start the next one. Finish it. Start the next one. Three features, 45 minutes. Sequential. Meanwhile, someone on Reddit shipped the same three features in 15 minutes by running them in parallel.

The difference isn’t talent. It’s four features that most Claude Code users don’t know exist. Google Trends confirms the wave: “claude dispatch” spiked 1,500% in a single week. “Claude code remote control” is up 550%.

Here’s what matters for you right now, what to ignore until later, and how to start this week.

What is remote control and how do you set it up?

Remote control connects your local Claude Code session to any browser or the Claude mobile app. From the official Claude Code docs: remote control lets you “Continue a local Claude Code session from your phone, tablet, or any browser.” Your code stays local. The phone is just a window into the running session.

Three ways to start:

# Dedicated server mode (terminal stays open, waits for connections)
claude remote-control

# Interactive session with remote access enabled
claude --remote-control

# From inside an existing session
/remote-control

Claude generates a session URL. Open it on your phone or scan the QR code (press spacebar in server mode to show it). You see what Claude is doing in real time. You approve or reject file changes. You send follow-up messages. You tell Claude to stop or change direction.

The session runs locally the entire time. Nothing moves to the cloud. Your filesystem, MCP servers, and project configuration all stay available.

Tips from real usage:

• Keep tasks self-contained. “Build a statistics dashboard showing task counts by status” works remotely. “Let’s brainstorm the dashboard layout” does not, because you are typing responses on a phone keyboard.
• Set effort to medium or low. Max effort tasks take longer and you are paying for time watching your phone.
• Review changes before approving. I have caught bugs in remote sessions that I would have missed at my desk.
• Use it for overnight tasks. Tell Claude to run your test suite, generate a coverage report, and list all untested functions. Check the report in the morning.

What does a real session look like with these features?

Before the details, here’s what your morning looks like once you adopt even two of these:

9:00 AM. Start Claude Code with remote control. From your phone while making coffee: “Run the test suite and tell me if anything is broken from yesterday.” Tests pass.

9:10 AM. Sit down. Four tasks for the morning. Three touch different files. Dispatch them in parallel. Go make a second coffee.

9:25 AM. Check phone. Two tasks clean, one has a minor bug. Approve the clean ones. Tell the buggy instance to fix the headers.

9:35 AM. Start the fourth task (it depended on the first one finishing).

10:30 AM. Four features built, tested, working. Without dispatch and remote control, that’s a 1 PM finish. You just bought back two and a half hours.

That’s the payoff. Now here’s how each piece works.

What is dispatch and how do you plan for it?

Dispatch sends multiple tasks to separate Claude Code instances that run in parallel. In the Desktop app, Dispatch is a persistent conversation that lives in the Cowork tab. You message it a task, and it decides how to handle it. Tasks that are development work get spawned as Code sessions automatically.

The constraint: dispatch works best when tasks touch different files. Two tasks editing the same file means merge conflicts.

# Three tasks, three instances, all running simultaneously
claude dispatch \
  "Add priority filtering to the task list" \
  "Write unit tests for the task CRUD operations" \
  "Add a loading skeleton to the main page"

Those three tasks work in parallel because filtering touches the UI component, tests touch the test files, and the skeleton touches the layout. No overlap.

The dispatch planning method:

1. List everything you want to build this session
2. Draw a dependency map (mental or on paper)
3. Identify which tasks are independent (different files)
4. Dispatch independent tasks in parallel
5. Queue dependent tasks for the next round

Real example. Four tasks for the morning:

1. Add a task statistics dashboard     → touches model + new component
2. Add email notifications for due     → new file, different feature
3. Write tests for the stats dashboard → DEPENDS ON task 1
4. Fix the mobile layout on task list  → touches layout component

Round 1 dispatch: tasks 1, 2, and 4 (independent)
Round 2 dispatch: task 3 (depends on task 1 finishing)

Result: 4 tasks in 2 rounds instead of 4 sequential.
         Total time cut roughly in half.

Start with two dispatched tasks, not five. Get used to reviewing parallel changes before scaling up.

What about channels and MCP?

Skip these for now. Channels and MCP are powerful but they’re for month 2, not week 1. Here’s the one-sentence version so you know what’s coming:

Channels push events from external systems (Telegram, Discord, CI pipelines) into your Claude Code session. Claude reacts without you opening a terminal. They’re in research preview.

MCP (Model Context Protocol) connects Claude to external tools like databases, browsers, and design files. It extends what Claude can read and write beyond your local filesystem.

Both are advanced. Focus on LSP, remote control, and dispatch first. When those feel natural, channels and MCP are the next level.

What does LSP actually change?

LSP (Language Server Protocol) drops code navigation from 30-60 seconds to 50 milliseconds. One Reddit post about enabling LSP in Claude Code got 862 upvotes on r/ClaudeCode. The claim is not exaggerated.

Without LSP, Claude finds a function by reading file after file. With LSP, Claude asks “where is deleteTask defined?” and gets an instant answer: src/app/actions.ts, line 47.

The token savings are massive. LSP reduces file reads by 60-80% on medium-sized projects. Every file Claude does not read is thousands of input tokens you do not pay for.

Setup is one line per language:

Language	Install command	Auto-detected?
TypeScript	npm install -D typescript (+ tsconfig.json)	Yes
Python	pip install pyright	Yes
Go	go install golang.org/x/tools/gopls@latest	Yes
Rust	Install rust-analyzer	Yes

Enable it now while your project is small. By the time your project grows, LSP will already be working.

Which feature should you use when?

Feature	What it does	Best for	Setup effort
LSP	Instant code navigation for Claude	Every session, every project	One-time install
Remote Control	Monitor/steer a local session from any device	Long tasks, overnight runs, away from desk	One flag: --remote-control
Dispatch	Parallel Claude instances on separate tasks	Independent features touching different files	Desktop app or CLI dispatch
Channels	Push external events into your session	CI failures, chat messages, webhooks	Plugin install + platform pairing
MCP	Connect Claude to external systems (DB, browser, Figma)	Querying production data, automated screenshots	Server config in .mcp.json
Voice Mode	Speak instructions instead of typing	Quick tasks, status checks, hands-free	/voice or settings toggle

The key distinction: remote control and dispatch are for work you initiate. Channels are for work that arrives from outside. LSP runs silently behind everything, making all of them faster.

What is the adoption timeline?

Do not adopt all of these at once. This order is based on impact-to-effort ratio:

When	Feature	What to do
Week 1	LSP	Install the language server. Forget about it. Zero learning curve, immediate benefit.
Week 2	Conversation discipline	Fresh conversations every 15-20 messages. Right effort levels. Lock in the habits before adding complexity.
Week 3	Remote control	Start a long task with claude --remote-control. Walk away. Approve from phone.
Week 4	Dispatch	Start with 2 parallel tasks. By end of week, try 3-4. Output doubles.
Week 5	Voice mode	Use for quick tasks and status checks. Do not force it on precise work.
Month 2	Channels + MCP	Advanced. Wait until basics are comfortable. Channels need a larger project.

The beginner shortcut: If you’re brand new, install LSP this week and ignore everything else on the table. Next week, try remote control on one long task. The week after, dispatch two tasks. That’s enough to double your output within a month.

What should you actually do?

• If you have never used any power feature: enable LSP right now. It takes one install command and pays for itself immediately with fewer token burns and faster responses.
• If you are comfortable with basic Claude Code: try remote control this week. Start one long task with claude --remote-control, walk away, approve from your phone. Once that feels natural, try dispatching two independent tasks in parallel.
• If you are already dispatching tasks: look at channels. A Telegram channel means you can message Claude from anywhere and get work done against your real local files. A webhook channel means CI failures trigger automatic investigation.

bottom_line

• You’re building features one at a time. Dispatch lets you build three in parallel. Remote control lets you approve changes from your phone. Together they cut build time in half.
• Start with LSP (one install, zero effort, immediate benefit). Add remote control week 2. Add dispatch week 3. Ignore channels and MCP until the basics feel natural.
• Dispatch planning takes 30 seconds: list tasks, check file overlap, dispatch the independent ones. That 30 seconds saves 30 minutes every session.