The shift from synchronous AI interactions to asynchronous AI partnerships just accelerated. Today, Anthropic announced Claude Code Channels, a feature that lets you message your running Claude Code session directly from Telegram or Discord. This is not just another chat interface. It represents a fundamental change in how developers interact with AI coding agents: from sitting at your terminal waiting for responses to having a persistent AI collaborator you can reach from anywhere.

Aspect	Key Point
What it is	MCP server that pushes events into Claude Code sessions
Supported platforms	Telegram and Discord (research preview)
Requirements	Claude Code v2.1.80+, Bun runtime, claude.ai account
Security model	Sender allowlist with pairing codes
Availability	Pro/Max users, Team/Enterprise requires admin opt-in

Why Channels Change the Game

Through implementing agentic AI systems across various projects, I have observed that the biggest friction in AI-assisted development is the requirement to stay at your terminal. You start a complex task, realize you need to step away, and either abandon the session or leave your machine running with an unattended AI.

Claude Code Channels solves this problem by decoupling your presence from your AI’s availability. A channel is an MCP server that pushes events into your running Claude Code session. You send a message from your phone, it arrives in your session, Claude processes it and replies back through the same channel.

The practical implications are significant. You can check on a long-running task from the coffee line. You can provide additional context when Claude hits a decision point. You can receive notifications when your AI completes work or encounters blockers. This matches the vision of AI agents as autonomous collaborators rather than tools you invoke and wait on.

How Channels Work

Channels operate through the Model Context Protocol (MCP), which I covered in depth in my MCP developer guide. When you launch Claude Code with the channels flag, it starts an MCP server that maintains a connection to your chosen messaging platform.

Events flow in both directions. Inbound messages from Telegram or Discord arrive in your terminal as channel events. Claude reads them, does the work, and calls the channel’s reply tool. The response appears on your messaging platform while your terminal shows the tool call and confirmation.

Key technical details matter here. Events only arrive while your session is open. For always-on setups, you run Claude in a background process or persistent terminal. When Claude replies through a channel, you see the inbound message in your terminal but not the outgoing reply text. This keeps your terminal uncluttered while maintaining full visibility into what Claude receives.

File transfers work as well. Telegram supports attachments up to 50MB, while Discord allows up to 10 files at 25MB each. This means you can send screenshots of bugs, reference documents, or any files Claude might need to complete its work.

Setting Up Telegram Integration

The Telegram setup requires the Bun runtime, which powers all official channel plugins. You create a bot through BotFather, install the plugin, configure your token, and restart Claude Code with the channels flag enabled.

The pairing process adds security. When you message your bot for the first time, it returns a pairing code. You enter this code in your Claude Code session to add your Telegram ID to the allowlist. After pairing, you switch the access policy to allowlist mode so strangers cannot obtain pairing codes by messaging your bot.

One important caveat: the official Telegram plugin specifically requires Bun. Attempts to run it with Node.js or Deno may cause unexpected errors. This is a deliberate design choice to leverage Bun’s performance characteristics for the polling operations channels require.

Setting Up Discord Integration

Discord setup follows a similar pattern but requires additional permissions configuration. You create an application in the Discord Developer Portal, enable Message Content Intent under Privileged Gateway Intents, then generate an invite URL with the necessary permissions.

Required permissions include View Channels, Send Messages, Send Messages in Threads, Read Message History, Attach Files, and Add Reactions. These permissions enable Claude to participate fully in your Discord workflow, including responding to threads and reacting to messages.

The pairing and allowlist process mirrors Telegram. DM your bot, receive a code, pair in your session, then lock down access. This dual-layer security ensures only approved users can push messages into your AI session.

The OpenClaw Context

Claude Code Channels arrives in a competitive context. OpenClaw, launched in November 2025, popularized the concept of AI assistants accessible through messaging apps. The project attracted a significant user base by offering Claude and other models through iMessage, Slack, Telegram, WhatsApp, and Discord.

The irony is not lost on the developer community. OpenClaw’s creator originally named his project “Clawd” until Anthropic sent a cease-and-desist for potential trademark violations. He was subsequently hired by OpenAI. Now Anthropic ships a direct competitor integrated into their flagship coding product.

The strategic difference matters for your AI coding tool selection. OpenClaw operates as a separate service that proxies to various AI providers. Claude Code Channels integrates directly with your running Claude Code session, meaning it has access to your full workspace context, file system, and development environment. When you message it, you are messaging an agent that already knows your codebase.

Security Considerations

Every approved channel plugin maintains a sender allowlist. Only IDs you have explicitly added can push messages. Everyone else gets silently dropped. This prevents random users from discovering your bot and sending commands to your AI.

The security model has three layers. First, being in your MCP configuration is not enough. A server must also be named in the channels flag to receive messages. Second, the pairing process ensures only devices you physically control can add themselves to the allowlist. Third, Team and Enterprise organizations can disable channels entirely until an admin explicitly enables them in managed settings.

Warning: If Claude hits a permission prompt while you are away from the terminal, the session pauses until you approve locally. For unattended use, there is a flag to bypass prompts, but only use it in environments you fully trust. Giving an AI unrestricted access to your system while you control it remotely carries obvious risks.

For teams implementing AI agent security practices, channels require explicit organizational opt-in. This prevents shadow IT scenarios where developers connect personal messaging accounts to company development environments without oversight.

Practical Use Cases

The most immediate value comes from monitoring long-running tasks. You start Claude on a complex refactoring, leave your desk, and check in via Telegram to see progress. When Claude needs clarification about a design decision, you provide context without returning to your workstation.

Build and test monitoring becomes more natural. You can configure channels to forward CI results so Claude can react to pipeline failures while you focus on other work. This moves toward the vision of production-ready AI systems that participate in the entire development lifecycle.

Remote pair programming also becomes feasible. A junior developer can work with Claude in a shared session while a senior engineer provides occasional guidance through Discord messages. The senior engineer does not need to context-switch fully into the codebase to provide useful input.

Current Limitations

Channels are in research preview, which means several constraints apply. The channels flag only accepts plugins from an Anthropic-maintained allowlist. Custom channels require a development flag that bypasses these restrictions. The protocol and syntax may change based on feedback during the preview period.

Console and API key authentication are not supported. You need a claude.ai login, which means channels currently work only for individual developers or teams on Anthropic’s consumer and business plans. This limitation may affect organizations with specific authentication requirements.

The two-platform launch with Telegram and Discord covers common developer use cases but excludes popular alternatives. Slack integration is notably absent despite being the dominant enterprise messaging platform. Anthropic indicates they will expand supported platforms based on feedback, but timeline remains unclear.

Frequently Asked Questions

Do I need to keep my computer running for channels to work?

Yes. Events only arrive while your Claude Code session is open. For always-on access, run Claude in a background process or persistent terminal. The channel connection requires an active session.

Can I use channels with multiple projects?

Each running Claude Code session has its own channel connection. If you want channels in multiple projects simultaneously, you run multiple sessions, each with its own bot configured.

What happens if someone unauthorized messages my bot?

If they are not on your allowlist, messages are silently dropped. They never reach your Claude Code session. The pairing code mechanism prevents unauthorized users from adding themselves while you are away.

Will channels work with Claude Code in CI/CD pipelines?

Currently no, since channels require claude.ai authentication. CI/CD typically uses API keys or service principals. This use case may expand as the feature matures.

Recommended Reading

• Agentic AI Foundation: MCP Developer Guide
• Agentic Coding: Transforming AI Engineering
• AI Coding Agent Production Safeguards
• AI Agent Development Practical Guide

Sources

• Push events into a running session with channels - Claude Code Documentation

---

To see exactly how to implement AI coding workflows in practice, watch the full tutorial on YouTube.

If you want to stay ahead of developments in AI coding tools and learn how to build production AI systems, join the AI Engineering community where we discuss tool selection, agentic workflows, and real implementation strategies.

Inside the community, you will find discussions on Claude Code optimization, AI agent development patterns, and guidance from engineers who have deployed these tools in production environments.