How to Use Claude Code Remote Control from Your Phone (2026)

Fastest official approach: Upgrade Claude Code to the current version, run claude remote-control in your project directory, then open claude.ai/code from the Claude App on your phone or from a browser to launch and control a session still running on your computer. If the session is already open, type /remote-control; if you also want to keep working in the local terminal at the same time, use claude --remote-control.

If you only use official Claude Code and a Claude.ai subscription, prefer the official Remote Control. If you use an API Key, Bedrock, Vertex, a custom ANTHROPIC_BASE_URL, or want to remotely control Codex, PandaCode, DeepSeek harness, and multiple development machines from the same interface, consider PandaNpc. This article explains both paths clearly, so you don't need to guess which one fits you first.

Information and commands were verified against the official Anthropic Remote Control documentation on August 30, 2026. Claude Code updates quickly; the actual available commands are still governed by your local claude --help and /help.

Three Ways to Start Claude Code Remote Control

The three commands are not synonyms; the difference is where you start from:

Official Remote Control execution still happens on your local machine: project files, MCP, tools, and configuration are not moved into a new cloud container; the phone and browser are only remote control interfaces. This is different from Claude Code on the web, which runs in an Anthropic-managed cloud environment.

Using Official Remote Control: From Checks to a Successful Connection

Step 1: Confirm Version, Login, and Workspace

First, on the development machine run:

claude --version

The official documentation requires Claude Code v2.1.51 or newer. Then start claude once as usual in the target project directory and complete /login and workspace trust confirmation. Remote Control uses the Claude.ai identity; the official docs currently state clearly that API Key is not supported. If Team or Enterprise organizations cannot see the entry point, an administrator also needs to enable Remote Control in the Claude Code admin settings.

Step 2: Choose a Startup Mode

To select a directory from your phone and start a new session, run:

cd /path/to/project
claude remote-control --name "My Project"

To keep typing in the development machine terminal while allowing your phone to take over, run:

cd /path/to/project
claude --remote-control "My Project"

For an already running session, type:

/remote-control My Project

The terminal shows the session URL; in claude remote-control mode, pressing space also displays a QR code.

Step 3: Connect from a Phone or Browser

Use the same Claude account:

• Open https://claude.ai/code in a browser and select the session with the computer icon in the session list;
• On iOS or Android, open the Claude App and go to the Code tab;
• Or open the URL the terminal gives you directly, or scan the QR code.

After the first connection, send a read-only verification prompt first; do not approve file writes immediately:

Print the current working directory and show git status --short. Do not modify anything.

Only when the path is correct, the Git status matches expectations, and permission confirmations appear on your phone or browser are you truly connected.

How to Choose Between Official Remote Control, Claude Code Web, and PandaNpc

So the choice is simple: if the official capabilities are enough, use the official solution. PandaNpc's value is not repackaging the same official command; it fills the gaps for cross-engine, API/custom model, unified multi-machine management, and private device network scenarios.

When the Official Option Doesn't Fit: Use PandaNpc to Remote-Control Your Local Agent

PandaNpc's PandaPaw runs alongside your development machine and registers local engines such as Claude Code and Codex as remote connections. Model calls still follow each engine's own account and provider configuration; the remote side only sends instructions, displays results, and handles approvals.

Step 1: Install and Log In to PandaPaw on the Development Machine

macOS / Linux:

curl -fsSL https://cos.pandanpc.com/pandapaw/install.sh | sh

Windows administrator PowerShell:

iwr -useb https://cos.pandanpc.com/pandapaw/install.ps1 | iex

The installer opens a browser authorization page. After approving, check:

pandapaw status

Step 2: Select the Host and Project from the Agent Page

Open PandaNpc Agent in a browser, confirm the target machine is green and online, select Claude Code and the correct project directory, then send the same read-only verification prompt as before. The phone and desktop clients share the same connection list.

If you only care about the permission cards, question cards, and Live Activity on iPhone, continue with the iOS workflow for connecting Claude Code from your phone.

Multiple Machines and Multiple Agents: Don't Do Port Mapping Machine by Machine

When you have only one development machine, a single remote session is enough; when Claude Code, Codex, and other Agents are spread across desktops, laptops, NAS, cloud servers, or dev containers, the real problem becomes host discovery and secure interconnection.

PandaNpc's connected devices joins the machines under your account into an encrypted private network, assigns each machine a stable private IP, and automatically registers the Agents it can run. Devices prefer peer-to-peer connections and fall back to relaying when hole punching fails; the whole process requires no public IP or router port mapping. This way, your phone only needs to choose "which machine, which Agent, which project" without storing a set of SSH ports, dynamic DNS names, and long-lived keys.

This kind of device networking solves "whether Agents can reach each other securely"; task decomposition, context sharing, and result aggregation still belong to the multi-Agent orchestration layer. The two must not be conflated.

Common Questions and Troubleshooting

`claude remote-control` or `/remote-control` Doesn't Exist

First check claude --version. The main command requires at least v2.1.51; VS Code support requires a newer version. After upgrading, reopen the terminal and session. If it still isn't there, run /help to confirm the organization policy hasn't disabled Remote Control.

Can't Find the Computer or Session on Your Phone

Confirm that the phone, browser, and CLI are logged in with the same Claude account and organization, and that the claude remote-control process on the development machine is still running. On the phone, go to the Code tab, not the normal chat list.

Can't Enable It with API Key, Bedrock, Vertex, or a Custom Gateway

Official Remote Control currently requires a Claude.ai login and explicitly does not support API Key. First confirm whether this is the authentication method you must keep; if you can switch back to a subscription login, use the official option; if not, use a solution that can connect to the local CLI, such as PandaNpc. Don't quietly change your production gateway or identity policy just to enable remote control.

Can the Computer Be Turned Off or Have Its Lid Closed?

No. The screen can be turned off, but the host running the Agent must stay powered on, the Claude/PandaPaw process must stay alive, and it must be able to reach the network. Official Remote Control reconnects automatically after the host recovers; if it sleeps frequently, switches networks, or shows "online but no longer sends messages," go through the Claude Code remote disconnection troubleshooting checklist item by item.

Will Multiple Remote Sessions Modify the Same File?

There is a conflict risk. Official server mode can create multiple sessions from the same directory by default; for parallel modification, use claude remote-control --spawn worktree to give sessions independent Git worktrees. In PandaNpc, you should also choose different workspaces or worktrees for parallel Agents; don't let two write-type sessions operate on the same checkout at the same time.

PandaNpc Shows the Machine as Offline

Run pandapaw status on the host and confirm, in order: account authorization, background service, outbound HTTPS/WebSocket, and the corresponding engine login. Opening a public inbound port cannot fix an outbound connection blocked by a corporate proxy.

Security Boundary: Code Stays Local, But That Doesn't Mean Nothing Travels Over the Network

Official Remote Control has the local machine actively establish an outbound HTTPS connection and does not open an inbound port on the router; messages travel between the remote interface and the local session over the Anthropic API's TLS connections. Project files and tool execution stay on the host, but to display the conversation and process instructions, session data still needs to pass through the official service.

PandaNpc likewise has the host actively connect to the authentication relay, with execution staying on the host; messages, status, diffs, and approvals that the remote side needs to display pass through the relay. Whichever solution you use, you should:

• Verify the host and working directory before approving file writes or running commands;
• Not share remote connections with untrusted people;
• For sensitive projects, prefer default approval or edit-only mode;
• On company devices, follow company-approved VPN, VDI, identity, and audit requirements.

FAQ

What is Claude Code Remote Control?

It lets the Claude App or claude.ai/code control Claude Code still running on your computer. The phone and browser are remote interfaces; local files, MCP, tools, and project configuration stay on the development machine.

How do I use Claude Code from my phone?

Run claude remote-control or claude --remote-control on the development machine, or type /remote-control in an existing session; then open the Code tab of the Claude App or claude.ai/code with the same account. If you also need to control Codex, custom models, or multiple hosts, you can use PandaNpc's Agent page.

Does Claude Code Remote Control Support Browsers?

Yes. The official entry point is claude.ai/code; PandaNpc also offers browser, iOS, and desktop clients. Both keep the Agent executing on the host, rather than copying the entire project to the current browser.

Does Claude Code Remote Control Support API Keys?

The official docs currently state clearly that API Key is not supported; you need to log in via Claude.ai. If your Claude Code must use an API, Bedrock, Vertex, or a custom ANTHROPIC_BASE_URL, keep your original authentication and choose a remote solution that can connect to the local CLI.

Will Claude Code Stop If My Phone Goes Offline?

As long as the development machine and the Agent process are still online, your phone going offline or closing the page does not stop the host session. You can continue viewing after reconnecting. However, no new work can be executed while the development machine is sleeping, powered off, or offline for an extended period.

Can Official Remote Control Control Codex or Multiple Agents?

No, it is a Claude Code feature. To manage Claude Code, Codex, PandaCode, DeepSeek harness, Pi, and multiple hosts from the same interface, you need an additional multi-engine connection layer; PandaNpc's PandaPaw and connected devices are designed for this scenario.

Conclusion

If you just want to continue an official Claude Code session from your phone or browser: run claude remote-control; the official option is the shortest and most direct. If you need API/custom models, other engines such as Codex, multiple hosts, or private device networking: keep your local Agent configuration and unify the connections in PandaNpc through PandaPaw.

Whichever path you choose, the success criterion is not "the page shows online" but that the remote side can read the correct working directory, see the real Git status, receive permission confirmations, and continue the same local session after a disconnect and reconnect.