Send a Telegram message, get a code fix back. That is the pitch behind Claude Code Channels, a new research preview feature that turns chat apps into remote controls for Claude Code sessions.
The feature works by running MCP servers (Model Context Protocol, Anthropic's standard for connecting AI tools to external services) as plugins inside Claude Code. Each plugin polls a Telegram or Discord bot for incoming messages, injects them into your active coding session, and lets Claude reply back through the same channel. You set up a bot on either platform, install the official plugin, and pair it with a one-time code.
The practical upside: you can fire off a message from your phone asking Claude to fix a failing test, refactor a module, or check on a running process, and it handles it in whatever terminal session you have open. No need to be sitting at your laptop.
How Setup Works
- Requires Claude Code v2.1.80 or later and the Bun JavaScript runtime
- Install with
/plugin install telegram@claude-plugins-officialor the Discord equivalent - Launch with
claude --channels plugin:telegram@claude-plugins-official - Create a bot via Telegram's BotFather or Discord's Developer Portal, save the token
- Pair using a one-time code exchange
Security gets a few guardrails. Every channel plugin maintains a sender allowlist, so only approved user IDs can push messages into your session. Channels only work with claude.ai login, not API keys. Team and Enterprise organizations have to explicitly enable the feature through admin settings.
One important limitation: events only arrive while a Claude Code session is running. There is no built-in daemon mode. If you want always-on availability, you need to keep a terminal open in something like tmux or screen. A headless remote-control mode is an open feature request but has not shipped yet.
The OpenClaw Comparison
This is clearly aimed at the crowd that was buying dedicated Mac Minis to run OpenClaw, a community-built tool for autonomous AI coding that went viral earlier this year. Channels removes the hardware requirement and wraps it in Anthropic's security model, but it is more limited in scope. OpenClaw was designed for fully autonomous long-running tasks. Channels is closer to a remote command interface for an existing session.
The feature is in research preview, meaning the --channels flag syntax and plugin contract could change. Only plugins from Anthropic's official allowlisted repository are supported during this phase.