Anthropic just recently released a walkie-talkie, and it's called Channels
I've been using Claude Code, Gemini, and ChatGPT mostly as a back and forth chat in their interface. When I typed something, it responded. This behaviour stays the same if I don't extend it's capabilities.
That’s where channels come in. They let outside apps like Telegram, Discord, CI pipelines, and alerts talk to a running Claude Code session. At a high level, a channel is just an MCP server that pipes events into a your session.
So, what does that actually look like?
Imagine you’ve integrated Telegram as a channel to Claude Code on your machine. Now you can text Claude from your phone while you’re outside getting your steps in.
The command to set it up looks something like this:
So, how does it work under the hood?
Channels is just an MCP. Each channel is basically an MCP server that sits between Claude and some external service. When something happens like a message, an alert, or build failures, the MCP server pushes it to Claude's session as an event.
With channels, the destination is a live Claude session that has context of your project. Like your files, git history, and previous messages. It's also interactive! Claude can ask follow-up questions back through the channel.
So, whats the catch?
It not an always on agent. It's not running forever on its own. It only work if your session is open. If you close the terminal, then it's gone.
You might also like
Make Your Own Claude Code
How to build your own CLI coding assistant inspired by Claude Code — from terminal UI to tool use to agentic loops.
BlogOn Building your own Lobster Trap
BlogBuild Your Own GREMLIN IN THE SHELL
A hands-on guide to building your own shell-based AI agent that haunts your terminal and gets things done.