Claude Code Remote Control: Run Local Sessions From Any Device

What if you could start a Claude Code session on your desktop, walk away, and pick it right back up from your phone? No VPN, no SSH tunnel, no exposed ports, and your code never leaves your machine.

That's exactly what Remote Control does. Announced in February 2026, it's one of the most significant additions to Claude Code yet -- turning your local terminal into a remotely accessible development environment that works from any device with a browser.

This guide covers everything: what Remote Control is, how its security model works, all three ways to start a session, MCP integration, and practical workflows you can set up today.

What Is Claude Code Remote Control?

Remote Control lets you access a Claude Code terminal session running on your local machine from any remote device -- your phone, tablet, a different computer, or any browser. The critical distinction: execution always stays local. Your remote device is just a window into the session running on your machine.

Everything you'd have in a normal Claude Code session remains available:

• Full local filesystem access
• All connected MCP servers
• Project configuration (CLAUDE.md, hooks, skills, plugins)
• Your entire tool environment

Think of it like screen-sharing for your terminal, except the interface is Claude's native chat UI and it works over a secure, pull-based connection through Anthropic's API.

How It Works: The Security Architecture

The first question everyone asks is: "Doesn't this expose my machine to the internet?" No -- and here's why.

Remote Control uses outbound-only HTTPS polling, not inbound connections:

1. Your local Claude Code process initiates outbound HTTPS connections to Anthropic's API
2. The process continuously polls the API for instructions from your remote device
3. No inbound ports are opened on your machine
4. All traffic travels over TLS (Transport Layer Security)
5. Credentials are short-lived, single-purpose, and expire independently

This is fundamentally different from traditional remote access tools like VNC or SSH. It's pull-based, not push-based, which eliminates the attack surface of exposed ports entirely.

What about permissions?

Remote Control does not support --dangerously-skip-permissions. Every action that requires approval in a normal session still requires approval when accessed remotely. You'll get the same permission prompts on your phone that you'd see in your terminal.

Three Ways to Start a Remote Control Session

Option 1: Dedicated Server Mode

The most powerful option. Starts a dedicated Remote Control server that can handle multiple concurrent sessions:

claude remote-control

This mode is ideal when you want your machine to serve as a persistent development server. Available flags:

Example with all flags:

claude remote-control --name "Backend API" --spawn worktree --capacity 8 --verbose

This starts a server named "Backend API" that creates isolated git worktrees for each new session, allows up to 8 concurrent sessions, and logs everything.

Option 2: Interactive Session with Remote Access

Already working in a Claude Code session and want to make it remotely accessible? Add the --remote-control flag:

claude --remote-control
# or with a custom name:
claude --remote-control "Debugging auth flow"

This gives you a full interactive terminal session that's also accessible from your remote devices. You get the best of both worlds -- local terminal control plus remote access.

Option 3: Enable Mid-Conversation

Already deep into a conversation and realize you need to step away? Enable Remote Control without losing your context:

Your entire conversation history is preserved, and the session immediately becomes accessible from your other devices.

Connecting From Another Device

Once Remote Control is active, you have three ways to connect:

1. Direct URL

Open the session URL displayed in your terminal at claude.ai/code. All active Remote Control sessions appear in the session list with a green status indicator.

2. QR Code

In server mode, press spacebar to display a QR code. Scan it with the Claude app on your phone for instant access -- no typing URLs on a tiny screen.

3. Session Browser

Browse all your sessions at claude.ai/code or in the Claude mobile app. Remote Control sessions show a green dot indicating they're online and ready.

MCP Integration: Your Tools Travel With You

One of the biggest advantages of Remote Control over cloud-based alternatives is that all your MCP servers stay available. Since execution remains local, every MCP connection you've configured continues to work.

This means you can access from your phone:

• Database servers (Postgres, MongoDB, Redis) running on localhost
• Git operations through the Git MCP server
• File system browsing through the Filesystem MCP server
• Docker management through the Docker MCP server
• Cloud provider tools (AWS, Vercel, Cloudflare) configured locally

How MCP Flows Through Remote Control

When you issue a command from your remote device, the flow is:

1. Remote device sends message through Anthropic's API
2. Local Claude Code process receives the message
3. Claude decides to use an MCP tool (e.g., query a database)
4. MCP call executes locally against your running MCP server
5. Result flows back through the API to your remote device

The MCP transport layer supports three protocols:

All MCP communication uses JSON-RPC 2.0 regardless of transport, ensuring standardized request/response format and error handling.

Real-World Workflows

On-Call Debugging From Your Phone

You get paged at 2 AM. Instead of dragging your laptop out:

# Already running on your work machine:
claude remote-control --name "Incident Response"

Open the Claude app on your phone, connect to the session, and start investigating. You have full access to logs, databases, and your codebase -- all executing on your work machine where everything is already set up.

Multi-Device Development

Start a feature at your desk, continue on the couch with your tablet:

claude --remote-control "Feature: user dashboard"

The session persists across device switches. If your laptop sleeps or the network drops briefly, the session auto-reconnects when connectivity returns (up to ~10 minutes of outage).

Team Pair Programming

Share a live session with a teammate for real-time collaboration:

claude remote-control --name "Pair: auth refactor" --spawn same-dir

Your teammate connects to the same session through their browser. They see the full conversation history and can contribute prompts while execution stays on your machine.

Concurrent Task Management

Run multiple independent tasks from your phone while away from your desk:

claude remote-control --name "Multi-tasker" --spawn worktree --capacity 4

The worktree spawn mode creates isolated git worktrees for each session, so parallel tasks don't interfere with each other. Kick off a refactor in one session, a bug fix in another, and a code review in a third -- all running locally, all controlled from your phone.

Configuration & Prerequisites

Requirements

• Claude Code v2.1.51 or later
• Pro, Max, Team, or Enterprise plan
• OAuth authentication via claude.ai (API keys are not supported)
• Workspace trust established in your project directory

Enabling Remote Control

For a single session: Use the CLI flags described above.

For all sessions globally:

claude
# Then in the session:
/config

Navigate to "Enable Remote Control for all sessions" and toggle it on. Every future session will automatically be remotely accessible.

Important: OAuth Required

Remote Control requires OAuth authentication. If you're using an ANTHROPIC_API_KEY environment variable, Remote Control will fail. Make sure you're authenticated with:

/login

If you're on a Team or Enterprise plan, your admin must also enable Remote Control in the organization settings at claude.ai/admin-settings/claude-code.

Remote Control vs. Claude Code on the Web

Anthropic now offers two ways to use Claude Code remotely. Here's how they compare:

Use Remote Control when you have an existing local environment with MCP servers, databases, and project configuration you need access to. Use Claude Code on the Web when you want to start fresh without any local setup.

Troubleshooting

Limitations to Know

• One remote session per interactive process. Use server mode (claude remote-control) with --spawn for multiple concurrent sessions.
• Terminal must stay open. Closing your terminal ends the session. Consider running in tmux or screen for persistence.
• ~10-minute reconnection window. If your machine loses connectivity for longer than ~10 minutes, the session terminates.
• Permission prompts still require approval. You can't bypass permissions remotely -- this is a deliberate security decision.
• No --dangerously-skip-permissions support. Every action must be explicitly approved.

Pro Tip: Use tmux for Persistent Sessions

To keep Remote Control running even when you close your laptop lid:

tmux new-session -s claude
claude remote-control --name "Always On"
# Detach with Ctrl+B then D

Your session survives terminal closures, laptop sleep, and SSH disconnections.

Getting Started Today

Remote Control is the fastest way to unlock mobile and multi-device development with Claude Code. Here's your quickstart:

1. Update Claude Code to v2.1.51 or later
2. Authenticate with OAuth: Run /login in a Claude Code session
3. Start a session: claude --remote-control "My first remote session"
4. Connect from your phone: Open claude.ai/code or the Claude app and look for the green dot
5. Start coding: Everything works exactly as it does locally

Combine Remote Control with hooks for automated quality checks, MCP servers for expanded tool access, and CLAUDE.md project instructions for a development environment that travels with you everywhere.

Using Remote Control in a creative way? Share your workflow with the Claude Directory community and help other developers work from anywhere.