Anthropic Just Shipped Claude Channels — Your AI Coding Agent Now Talks to Telegram and Discord

Anthropic just dropped one of the most interesting features I have seen in the AI coding space this year. It is called Channels, and it fundamentally changes how you interact with Claude Code.

The short version: you can now push messages into a running Claude Code session from external platforms like Telegram, Discord, webhooks, CI pipelines, and monitoring tools. Claude reads the message, does the work, and replies — all while you are away from your terminal.

What Are Claude Channels?

A channel is an MCP (Model Context Protocol) server that pushes events into your running Claude Code session. Unlike traditional integrations where Claude polls for data or you manually paste context, channels deliver information to Claude in real time.

Think of it like giving Claude a phone. Your CI pipeline fails? Claude gets notified instantly. Someone DMs your Telegram bot with a question about the codebase? Claude reads it, checks the files, and replies through Telegram. A monitoring alert fires? Claude sees it and starts investigating — all in the same session where it already has your project context loaded.

Telegram and Discord Are First

The research preview ships with official support for Telegram and Discord as channel plugins, plus a localhost demo called "fakechat" for testing. The setup is straightforward:

For Telegram, you create a bot through BotFather, install the plugin in Claude Code with /plugin install telegram@claude-plugins-official, configure your token, and restart with the --channels flag. Then you pair your Telegram account by DMing your bot, getting a pairing code, and approving it in Claude Code. From that point on, anything you send your bot goes straight to Claude.

Discord follows the same pattern — create a bot in the Developer Portal, install the plugin, configure, and pair. Both channels support a sender allowlist so only approved users can push messages into your session.

Two-Way Communication Is the Real Power

What makes channels genuinely useful is that they are not just one-way notification pipes. Claude can reply back through the same channel. Send your Telegram bot "what changed in the last commit?" from your phone, and Claude checks git log, reads the diffs, and sends you a summary right there in Telegram.

This is different from Anthropic's other remote features. Claude Code on the web runs tasks in a fresh cloud sandbox. Claude in Slack spawns a web session from a mention. Remote Control lets you steer a session from claude.ai. Channels fill the gap — they push events from non-Claude sources into your already-running local session, where Claude has full access to your real files and project context.

Webhooks Open Up Serious Automation

The channels reference documentation includes a full guide for building a custom webhook receiver. The concept is simple: you write a small MCP server (about 30 lines of TypeScript) that listens on a local HTTP port, and anything that POSTs to it gets forwarded to Claude as a channel event.

The practical applications here are significant. Your CI pipeline can POST a failure notification and Claude immediately starts investigating the broken build — in the same session where it was already working on your code. An error tracker can push alerts and Claude triages them. A deploy pipeline can notify Claude when a release goes out.

Channels fill the gap by pushing events from non-Claude sources into your already-running local session — where Claude has your files open and remembers what you were debugging.

Security Is Handled Sensibly

Every channel maintains a sender allowlist. Only IDs you have explicitly approved can push messages — everyone else is silently dropped. The pairing flow for Telegram and Discord requires you to approve a code in your Claude Code terminal before any messages get through.

Being listed in your MCP config is not enough either. A server also has to be named in the --channels flag when you start Claude Code. And on Team and Enterprise plans, an admin has to explicitly enable channels in managed settings before anyone on the team can use them.

There is also a permission relay feature. When Claude wants to run a tool that needs your approval, the prompt can be forwarded to your Telegram or Discord. You approve or deny from your phone, and Claude continues. The local terminal dialog stays open too — whichever answer arrives first wins.

Building Your Own Channel

If you need a channel for a system that does not have a plugin yet, you can build one. The requirements are minimal: declare the claude/channel capability in your MCP server, emit notifications/claude/channel events, and connect over stdio. Anthropic provides the full MCP SDK and detailed documentation for the contract.

During the research preview, custom channels need the --dangerously-load-development-channels flag to run, since they are not on the approved allowlist yet. But the architecture is open and well-documented — I expect we will see community channels for Slack, Teams, PagerDuty, and GitHub Actions fairly quickly.

What This Means for the Industry

This is Anthropic pushing Claude Code from "a really good AI pair programmer" toward "an autonomous agent that participates in your development workflow." The difference is subtle but important. An AI that waits for you to type in a terminal is useful. An AI that receives alerts, investigates issues, and reports back through your existing communication tools is a team member.

For businesses running development teams, this changes the economics. A Claude Code session running with channels can monitor CI, respond to team questions, and triage errors — all while your developers focus on higher-value work. It is not replacing anyone. It is handling the interrupt-driven work that fragments developer attention.

Requirements and Availability

Channels require Claude Code v2.1.80 or later and a claude.ai login (Console and API key auth are not supported yet). You need Bun installed to run the pre-built plugins. Team and Enterprise orgs need an admin to enable channels in managed settings.

This is a research preview, so expect the --channels flag syntax and protocol to evolve. But the foundation is solid and the documentation is thorough — Anthropic clearly intends this to become a core part of the Claude Code experience.

Worth Paying Attention To

If you are a developer or run a team that uses Claude Code, channels are worth setting up now. Start with the fakechat demo to understand the flow, then connect Telegram or Discord for real-world use. If you are building internal tooling, the webhook receiver pattern is immediately actionable.

At Tally Digital, we are already exploring how channels fit into client workflows — especially for teams that want AI-assisted monitoring and incident response without switching context. If you are curious about integrating AI agents into your development process, book a call and let us walk you through what is possible.