Remote Control is a feature of Claude Code that enables users to continue a local Claude Code session from any device, including phones, tablets, or other web browsers. Its primary purpose is to allow users to start a task on their local machine and seamlessly pick it up on a remote device, while ensuring that the underlying Claude Code session, including its computational processes, filesystem interactions, and tool executions, continues to run locally on the user's machine. This means no user data, code, or environment context is moved to the cloud.

The core methodology of Remote Control is based on a secure, outbound-only connection model. When a Remote Control session is initiated on the user's local machine, it registers with the Anthropic API and then actively polls for incoming work or commands from remote clients. The local Claude Code process only makes outbound HTTPS requests and does not open any inbound network ports on the user's machine, enhancing security. When a user connects from a remote device (e.g., via claude.ai/code or the Claude mobile app), the Anthropic API acts as a secure intermediary, routing messages and interactions between the remote web/mobile client and the locally running Claude Code session over a streaming connection. All traffic is encrypted using Transport Layer Security (TLS), and the connection utilizes multiple short-lived, purpose-scoped credentials that expire independently, further bolstering security. This architecture allows the user's full local environment—including filesystem access, Multi-Component Processor (MCP) servers, local tools, and project configurations—to remain available and interactable through the remote interface. The conversation and session state are kept in sync across all connected devices, allowing for interchangeable use of the terminal, browser, or mobile app.

To use Remote Control, a user must have a Pro or Max subscription (API keys are not supported), authenticate their Claude account via /login, and accept the workspace trust dialog by running claude in their project directory at least once.

Remote Control sessions can be initiated in two primary ways:

1. Starting a new session: By navigating to the project directory and executing claude remote-control. This command keeps the process running in the terminal, displays a session URL, and can show a QR code for quick mobile access. It supports flags like --verbose for detailed logs and --sandbox or --no-sandbox to enable or disable filesystem and network isolation.
2. From an existing session: By running the /remote-control (or /rc) command within an already active Claude Code session. This carries over the current conversation history and provides a session URL and QR code. This method does not support the sandboxing or verbose flags.

Once a session is active, remote devices can connect by opening the provided session URL in a web browser, scanning the QR code with the Claude mobile app, or finding the session by its name in the session list on claude.ai/code or the app. Remote Control sessions are identified by a computer icon with a green status dot when online. Users can also configure Claude Code to automatically enable Remote Control for all sessions by setting "Enable Remote Control for all sessions" to true via the /config command.

Remote Control fundamentally differs from "Claude Code on the web." While both use the claude.ai/code interface, Remote Control executes the session directly on the user's machine, leveraging local resources and configurations. In contrast, Claude Code on the web runs on Anthropic-managed cloud infrastructure, meaning it does not access the user's local environment. This distinction makes Remote Control suitable for continuing local work remotely, whereas Claude Code on the web is for initiating tasks without local setup or managing multiple parallel cloud-based tasks.

Limitations of Remote Control include:

• Only one remote connection is supported per active Claude Code instance.
• The terminal running claude remote-control must remain open; closing it or stopping the process ends the session.
• An extended network outage (approximately 10 minutes) while the machine is awake can cause the session to time out and the process to exit.