Claude Code Channels Permission Relay: Approve Tool Prompts from Your Phone

Claude Code

Claude Code 2.1.81 extends the Channels feature with a permission relay capability: channel servers that declare the permission capability can now forward tool approval prompts to the user's phone rather than pausing the session until someone returns to the terminal. This closes the main gap in unattended agentic workflows β€” a pending permission prompt no longer blocks a remote session indefinitely. The feature builds on the Channels research preview (introduced in v2.1.80) and requires a claude.ai login to operate.


Closing the Last Gap in Unattended Agentic Workflows

When Anthropic launched Channels in Claude Code 2.1.80, the premise was clear: developers could push messages from Telegram and Discord directly into a running Claude Code session, letting the AI react to CI builds, monitoring alerts, and chat instructions while the developer was away from the terminal. One problem remained β€” if Claude encountered a tool that required user approval, the session would pause and wait indefinitely.

Version 2.1.81 addresses this with the permission relay.

How the Permission Relay Works

Channel servers that declare the permission capability can intercept tool approval prompts and forward them to the user's phone through the same messaging channel. Rather than the session stalling on a permission dialog that no one is watching, the approval request travels through the channel β€” arriving as a message on Telegram, Discord, or whichever platform the channel server supports.

The user receives the prompt, approves or denies it remotely, and the session continues. No terminal presence required.

Why This Matters

The Channels feature transforms Claude Code from a request-response tool into an event-driven agent. But an always-on agent that can be interrupted indefinitely by permission prompts is not truly autonomous. The permission relay addresses the core tension between Claude Code's safety model β€” requiring explicit approval for potentially destructive operations β€” and the goal of letting the agent work unattended for extended periods.

Developers running long refactoring tasks, automated testing cycles, or continuous integration workflows can now receive and respond to permission requests from their phone, keeping sessions unblocked without weakening the underlying safety model. Crucially, the relay routes prompts to a human for approval rather than auto-approving them β€” the safety contract remains intact.

Context: Channels Research Preview

Channels remain in research preview as of v2.1.81. The feature requires Claude Code v2.1.80 or later and a claude.ai login; API key authentication is not supported. Telegram and Discord are the current supported platforms, with the plugin architecture designed for community expansion. Teams and Enterprise organizations must explicitly enable Channels in their settings.

The permission relay is an additive capability β€” existing Channels setups do not automatically gain it. The channel server plugin must explicitly declare the permission capability for relay to activate.

Claude Code Channels: Approve Tool Prompts Remotely from Your Phone | Yet Another Changelog