Claude Code — Remote Control

Drive a Claude Code session running on your local machine from any browser (claude.ai/code) or the Claude mobile app (iOS/Android). Your code, filesystem, MCP servers, and tools all stay local — the web/mobile UI is just a window into that local session.

Why / When to Use

When you want to continue a local Claude Code session from another device (phone, tablet, second laptop) without re-uploading files or losing context. Also useful for monitoring long-running tasks and receiving push notifications when Claude needs input.

Core Concept / Commands

Minimum version

claude --version   # must be v2.1.51+

Start a Remote Control session

Connect from another device: open the session URL, scan the QR code, or find the session by name at claude.ai/code.

Push notifications (v2.1.110+)

Claude pings your phone when a task finishes or needs input. Requires the Claude mobile app to be installed.

Key Options / Variants

Gotchas

• The local process must keep running — closing the terminal ends the session.
• Network outage > ~10 min causes timeout.
• Starting an ultraplan disconnects Remote Control.
• Some CLI commands (/mcp, /resume) are local-only and don’t work via remote.
• Security model: outbound HTTPS only, no inbound ports opened; short-lived scoped credentials via Anthropic API over TLS.

Source

Conversation: “CC-Remote Control” — 2026-05-16

Updates — 2026-05-17

Always-on Remote Control (built-in setting)

No alias needed — use the built-in config:

/config → enable "Enable Remote Control for all sessions"

This auto-starts Remote Control every time Claude Code launches interactively.

Alternatively, an alias:

alias cc='claude --remote-control'

Downsides of always-on Remote Control

• Startup overhead — registers with Anthropic API on every launch, adds ~1–2 seconds
• Internet required — if connection is flaky, Claude Code itself may fail to start
• Outage exits process — if machine loses network for ~10 min, the entire session dies
• Session accumulation — each terminal window registers its own remote session; you’ll accumulate sessions in the list on your phone
• Security surface — your local session is always reachable via Anthropic’s servers while running (TLS + scoped credentials, but still a consideration)

For local LLM dev work, the network dependency is the biggest practical risk — if your LiteLLM proxy or local network has issues, always-on Remote Control adds one more failure point on startup.

Updates — 2026-05-26

Why API Keys Don’t Work for Remote Control or Channels

API keys are inference-only — they authorise a request to the model and nothing more. Remote Control and Channels require a full identity layer that only OAuth login (claude auth login) provides.

An API key says “this request is billable to account X”. OAuth says “this is a real human user with a subscription, a phone, an identity”. Remote Control and Channels are built on top of that identity layer.

Local proxy users: If your API key points to a local LiteLLM/DeepSeek proxy (e.g. 172.18.0.1:4001), Remote Control won’t work even with a real Anthropic key — because session registration must go through claude.ai. To use Remote Control you must: run claude auth login, authenticate with your claude.ai account, and let inference go through Anthropic’s actual API (not a local proxy) for those features.

Source: Conversation “CC-Remote Control” — 2026-05-26

Updates — 2026-06-01

Third-Party Remote Control Tools

For local LiteLLM proxy setups where Official Remote Control won’t work (API-key-only + local proxy), third-party tools bridge Claude Code to chat platforms directly.

For LiteLLM proxy + Discord: OpenACP is the recommended first try — npm install -g @openacp/cli && openacp starts a wizard (choose platform, enter bot token, select workspace, choose agent).

Source: Conversation “CC-Remote Control” — 2026-06-01

Weekly Summary — 2026-W22

Appeared: Tue 26 May Key developments this week:

• Clarified the fundamental reason why API keys cannot activate Remote Control or Channels: API keys are inference-only credentials; they authorise a model request but provide no identity layer. Remote Control requires OAuth login (claude auth login) to register a session, verify identity, and enable push notifications.
• Documented the incompatibility with local proxy setups: even if ANTHROPIC_API_KEY is valid, pointing it at a LiteLLM/DeepSeek proxy makes Remote Control impossible because session registration must go through claude.ai.

New things learned:

• The distinction is inference-only (API key) vs identity+session layer (OAuth). These are architectural differences, not permission levels.
• To use Remote Control alongside a proxy setup: authenticate via claude auth login for session registration, and accept that inference for Remote Control features will route through Anthropic’s actual API.

Open questions / next steps:

• No new tasks — this was a conceptual clarification, no action required

Updates — 2026-05-31

Community Alternatives for Messaging Bot Control (Telegram/Discord)

Landscape of tools for sending commands to Claude Code via messaging apps when claude.ai OAuth is unavailable (e.g. local proxy setups):

Recommendation for local proxy users: Start with Claude-Code-Remote — purpose-built for the same workflow. OpenClaw is powerful but overkill for simple command dispatch.

Source: Conversation “CC-Remote Control” — 2026-05-31