Anthropic Just Dropped the Feature Everyone Asked For

Detailed Summary

Introduction (0:00 - 0:34)

Anthropic released a feature that allows users to control Claude Code sessions running on a local machine from a phone or another device. While this was previously possible using tools like tmux, Tailscale, and Terminus, the new native integration simplifies the process into a messaging-app-like experience within the Claude app.

Enabling Remote Control (0:34 - 2:46)

To start a remote session, users navigate to their desired directory and run the command claude rc or claude remote-control.

• The first run requires enabling a secure connection to Anthropic's servers.
• Users can enable remote control globally for all future sessions via /config.
• The interface allows users to switch between different active sessions, such as a video editing project or a software development task.
• Permissions can be granted directly from the phone, making it less 'fiddly' than previous manual setups.

Benefits and MCP Integration (2:46 - 3:37)

Remote control provides full access to local configurations, including custom skills and MCP servers.

• Users can trigger 'Plan Mode' or specific servers like Exa for background research from their mobile device.
• This allows for interactive planning where the AI asks questions that the user can answer while away from their desk, such as during lunch.

Security and Remote Servers (3:37 - 5:58)

Security is a primary concern when allowing remote access to a file system.

• Sessions are securely routed through Anthropic's infrastructure.
• To mitigate risks like prompt injection or accidental file deletion, users can host Claude Code on a remote VPS (e.g., Hetzner for ~$3.50/month).
• By running Claude Code in a tmux session on a remote server, users create a persistent agent that can perform web research and tasks without risking their local hardware or requiring it to stay online.

Sandboxing and Proxies (5:58 - 8:28)

For users running sessions locally, sandboxing and proxies offer granular control.

• Sandboxing: Users can enable a sandbox in settings.local.json to restrict Claude to specific directories, allow only certain websites, and deny dangerous tools like bash or file writing.
• Proxies: A custom Node.js proxy can be used to whitelist specific domains (like api.anthropic.com for the connection and research sites) while blocking all other traffic.
• These methods allow for 'dangerously skip permissions' mode to be used more safely by defining strict boundaries for the AI's behavior.

Differences and Future Outlook (8:28 - 9:09)

Claude Code's remote feature is currently more passive compared to tools like OpenClaw; it performs tasks when prompted rather than proactively searching on a schedule.

• A highly requested future feature is the integration of mobile notifications via hooks to alert users when a task is complete or requires input.
• The rapid pace of updates from Anthropic suggests proactive features may be added soon.

Conclusion (09:09 - 09:56)

The remote control feature may replace the need for third-party wrappers for many users. It allows for a centralized 'OS' folder (like RayOS) to be managed entirely from a phone, handling emails, file downloads, and video editing efficiently using a single Claude Code subscription.

Notable Quotes

"This new feature is being able to control Claude Code running locally on your machine from your phone or another device." — Ray Amjad

"If I was kind of worried about it accidentally being prompt injected... I could either use the Claude Code sandbox or have Claude Code running on a remote server." — Ray Amjad

"The fact that I can now use this remotely from my phone means that I can just have it running all the time and then just be giving prompts to my RayOS folder." — Ray Amjad