Claude Remote Control: use Claude Code from any device

Summary

Update Claude Code, sign in with an eligible claude.ai account, trust the workspace locally, and confirm any organization setting. Start server mode, an interactive remote-enabled session, or Remote Control inside an existing session. Connect through the generated URL, QR code, web session list, or mobile app. Keep the computer online, review permissions on either surface, and end the local process when remote access is no longer needed.

What Claude Remote Control does

The Claude Code remote interface is a synchronized window into a local process. Messages can be sent from the terminal, browser, or phone, and the conversation stays synchronized. Local file paths remain available through supported autocomplete. If the computer sleeps or loses its network, the connection can resume when the machine returns. It cannot keep executing while the local process and machine are unavailable.

What it is not

• Remote desktop shows the full computer display; this feature shows the Claude conversation.
• Claude Code on the web runs the coding task in an Anthropic-managed environment.
• Teleport continues a compatible cloud session on the local machine, which is the opposite direction.
• Cowork Dispatch controls Claude Desktop and Cowork workflows rather than this local Claude Code session.
• Remote access does not replace tool permissions, workspace trust, sandboxing, source review, or local operating-system security.

Check current requirements

Claude Code version

the current CLI documentation requires version 2.1.51 or later. The VS Code command requires 2.1.79 or later.

Plan

the current eligible plans are Pro, Max, Team, and Enterprise. Preview access may change.

Authentication

sign in through claude.ai. API keys and supported third-party provider modes do not work for Remote Control.

Organization

Team and Enterprise administrators must enable the feature because it is off by default for those plans.

Workspace

open the project in Claude Code locally and accept the workspace trust dialog before remote use.

Connectivity

the local computer, local Claude Code process, Anthropic service, and remote device need working network access.

Run `claude --version` and `/status` before troubleshooting. An environment variable that forces an API key, cloud provider, or disabled nonessential traffic can make an otherwise eligible account appear unavailable. Use the official error section to match the exact cause instead of changing unrelated settings.

Check the current Remote Control requirements and commands in Anthropic's documentation.

Start modes

Server

Run `claude remote-control` from the project directory when the terminal should wait for remote connections. This mode can create sessions from the remote surface. The process displays a session URL, and the spacebar can show a QR code. Review the current concurrency, naming, verbose logging, and sandbox options before using this mode. Sandboxing is currently off by default unless enabled with the documented flag or configuration.

Interactive terminal mode

Run `claude --remote-control` or `claude --rc` when you want a normal local conversation that is also available remotely. You can pass a session name. The terminal and remote surfaces remain active, so a message or approval can come from either one.

Existing session

Inside a current session, run `/remote-control` or `/rc` to carry its conversation history into remote access. A name can follow the CLI command. The in-session command does not accept the server-mode sandbox or verbose flags, so establish the required local controls before exposing the session.

VS Code extension

Use `/remote-control` or `/rc` in the extension prompt. The banner shows connection status and can open the browser session. The extension derives the title from the conversation rather than accepting a name argument. It also does not show a QR code in the same way as the terminal.

Connect from a browser or phone

1. 1Keep the local process running and note the session name, URL, and expected project.
2. 2On the remote device, sign in with the same account and organization used by the local Claude Code session.
3. 3Open the generated URL, scan the QR code in the Claude mobile app, or find the online computer-marked session in claude.ai/code or the app's Code section.
4. 4Confirm the session title and recent conversation before sending a prompt or approving a tool request.
5. 5Start with a read-only request and verify that the expected local project and tools are visible.
6. 6For a write task, inspect the proposed action, approve only the required permission, and review the local diff and test result.

Set an explicit session name when several local processes or repositories are active. A generated name can include a machine hostname prefix, while a conversation-derived title can change after the first prompt. Avoid placing a customer name, incident detail, secret, or other sensitive text in the title.

Understand the connection and security boundary

Anthropic says the local session makes outbound HTTPS requests and does not open an inbound port. It registers with the Anthropic API and exchanges messages through a streaming connection protected by TLS. The design uses several short-lived credentials with separate purposes. This reduces the need for direct inbound network exposure, but remote access still expands where a person can instruct and approve a locally powerful process.

Local scope

everything the local Claude Code process can reach can matter remotely, including files, credentials, tools, MCP servers, and network destinations.

Account scope

protect the claude.ai account and every connected device with current authentication and device-security controls.

Workspace scope

start the process only in the intended repository and remove unrelated directories, secrets, and production data from reach.

Permission scope

keep narrow tool rules and human approval. A remote surface is not a reason to enable blanket bypass mode.

Sandbox scope

verify whether a sandbox is active. Server mode currently exposes explicit sandbox flags and documents it as off by default.

Session scope

use a clear name, review connected devices, avoid sharing the URL or QR image, and stop the process when finished.

Recovery scope

keep version control, checkpoints, backups, cancellation, logs, and an incident path for unexpected actions.

Local execution does not mean that no data travels through Anthropic services. Claude Code still sends the messages and necessary task context required for the model and synchronized conversation, subject to the applicable product terms and controls. The distinction is that commands and filesystem operations execute on the local machine rather than a remote coding environment.

Remote Control or Claude Code on the web?

• Choose Remote Control to continue a live local session with its local files, tools, MCP servers, configuration, and environment.
• Choose Claude Code on the web for a task without local setup, an uncloned repository, or execution in Anthropic-managed infrastructure.
• Use teleport when a compatible cloud session should continue on the local machine.
• Use an ordinary remote desktop or secure shell only when you need the full machine interface or terminal rather than a synchronized Claude conversation.

Read Anthropic's current Claude Code on the web guide before choosing the execution location.

Set up mobile notifications carefully

Current mobile push support requires a newer Claude Code version than basic Remote Control. Install the Claude app, use the same account and organization, allow operating-system notifications, then enable the documented push setting in `/config`. Claude decides when to send a push, and there is not currently a full per-event rule set. Treat a notification as a prompt to inspect the session, not as proof that work succeeded.

Troubleshoot in the right order

1. 1Run `claude --version` and compare it with the current requirement for the exact surface and feature.
2. 2Run `/status` and confirm claude.ai authentication, the expected account, organization, and eligible plan.
3. 3For Team or Enterprise, ask the administrator to confirm the current Remote Control toggle.
4. 4Check for environment variables that disable nonessential traffic, set an API key, or force Bedrock, Vertex, or Foundry.
5. 5Confirm workspace trust locally and start a fresh named session from the intended directory.
6. 6Verify outbound HTTPS access and check the current Anthropic service status without opening an inbound port.
7. 7Sign out and back in only when the official error guidance identifies a stale eligibility or authentication state.
8. 8Use verbose logs in a supported start mode, remove sensitive values, and share the smallest reproducible error with support.

End a remote session safely

• Review the final conversation, tool calls, file diff, tests, unresolved questions, and external side effects.
• Cancel or roll back unsuitable changes with the project's normal recovery process.
• Disconnect the remote surface and stop the local Claude Code process when continued access is not required.
• Close sensitive browser tabs, secure the phone or tablet, and revoke a device or account session if it was lost or shared.
• Remove temporary credentials, files, tunnels, allow rules, and elevated access created for the task.
• Record any incident and update the team's remote-work policy before the next session.

What current remote guides emphasize

Current tutorials emphasize setup, mobile connection, terminal modes, local environment access, web comparison, push notifications, and common errors. This guide keeps those tasks and adds explicit execution, account, workspace, permission, sandbox, session, data-flow, and recovery boundaries. It also separates Remote Control from teleport, Dispatch, remote desktop, and secure shell.

Install and verify Claude Code locally before enabling remote access.

Compare other Claude Code interfaces and choose the correct execution surface.

Frequently asked questions