Anthropic’s Claude Code Channels Bring Always‑On AI Coding to Telegram and Discord

Anthropic is taking a direct swing at the wave of always-on, chat-based coding agents popularized by OpenClaw. With the release of Claude Code Channels, the company is turning its Claude Code environment into a persistent AI teammate you can message from Telegram and Discord—no dedicated Mac mini, complex self-hosting, or flaky bridges required.

For developers and AI power users, this release essentially brings core OpenClaw-style behavior—24/7 access to an autonomous agent over common messaging apps—into Anthropic’s official toolchain, with an emphasis on safety, managed infrastructure, and a standardized integration layer.

From desktop-only harness to always-on chat agent

Claude Code started life as an “agentic harness” around Anthropic’s Claude models: a way to write, run, test, and debug code interactively, with the AI calling tools, executing commands, and iterating in a loop. Until now, that experience was largely tied to:

• the Claude desktop app
• terminal or supported developer environments
• the Claude mobile app using a Remote Control setting that, in practice, could be unreliable

Claude Code Channels changes that interaction pattern. Instead of a purely synchronous, “ask-and-wait” chat interface, your Claude Code session can now sit in the background—on a terminal, server, or VPS—and wake up whenever it receives a message from Telegram or Discord.

This is the same interaction model that made the open source project OpenClaw explode in popularity after its November 2025 release by Austrian developer Peter Steinberger. OpenClaw let users treat an AI as a persistent worker reachable over iMessage, Slack, Telegram, WhatsApp, and Discord, capable of doing real work: drafting emails, organizing files, generating full applications, applying to jobs, or running long-running marketing campaigns—and then proactively messaging the user when tasks completed.

Anthropic is now offering a similar “persistent worker over chat” experience, but integrated into Claude Code and backed by the company’s own safety-focused brand and infrastructure. For many developers who found OpenClaw powerful but intimidating or risky, this lowers the barrier to trying an always-on agent.

Why OpenClaw mattered—and where it struggled

OpenClaw’s key innovation wasn’t just autonomous behavior; it was availability. Users could message their agent at any hour over familiar apps and get real work done with minimal friction. That “text your AI employee” model resonated with both professional developers and so-called “vibe coders” experimenting with side projects.

But OpenClaw also highlighted the tradeoffs of powerful, local-first agents:

• Security exposure: When you give an agent access to your file system, email, or other personal data, it can “run amok” if misconfigured or if a prompt goes wrong. That risk spurred an ecosystem of derivatives like NanoClaw, KiloClaw, and Nvidia’s NemoClaw, each promising better sandboxing and enterprise controls.
• Operational complexity: Running a 24/7 OpenClaw agent often meant buying dedicated hardware (like a Mac mini), configuring the stack, and maintaining it. For non‑technical users—or even busy engineers—that overhead was significant.

Claude Code Channels offers a different value proposition: similar always-on, chat-based access, but delivered as part of a managed, proprietary environment designed to be easier to set up, especially for those already using Claude Code.

How Claude Code Channels actually works

Under the hood, Channels is built on the Model Context Protocol (MCP), an open standard Anthropic introduced in 2024 to normalize how AI models connect to tools and data. MCP serves as the “universal port” between Claude and outside services.

In the Channels architecture:

• You start a Claude Code session with the --channels flag.
• That session spins up a background polling service using the Bun JavaScript runtime, known for its high-speed execution.
• Official MCP plugins for Telegram and Discord act as bridges. When a new message hits your bot, the plugin injects it into your active Claude Code session as a <channel> event.
• Claude Code then uses its existing tools—running code, tests, or fixes—to respond to the request, and sends the result back out via a dedicated reply tool.

The crucial difference from a typical web chat is persistence. A Claude Code session can sit in a terminal window or on a persistent server, quietly listening. When you ping it via Telegram or Discord, it can resume work, run tasks, and message you when it’s done, without you manually re‑opening a browser or re‑establishing context.

Step-by-step: connecting Telegram and Discord

To use Channels, you’ll need Claude Code v2.1.80 or later and the Bun runtime installed on your Mac or PC. The connectors rely on Claude’s official MCP plugins for Telegram and Discord.

Telegram setup flow

1. Create a Telegram bot. In Telegram, open BotFather and run /newbot to get a bot and access token.
2. Install the plugin. In your Claude Code terminal, run:
   /plugin install telegram@claude-plugins-official
3. Configure credentials. Save your token by running:
   /telegram:configure <your-token>
4. Launch with channels. Restart Claude Code with:
   claude --channels plugin:telegram@claude-plugins-official
5. Pair your account. DM your new bot in Telegram, get the pairing code it sends you, then enter:
   /telegram:access pair <code>

Discord setup flow

1. Create an application. In the Discord Developer Portal, create a “New Application” and grab the bot token (reset it if needed).
2. Enable privileged intents. In the Bot settings for that application, turn on Message Content Intent under “Privileged Gateway Intents” so the bot can see your messages.
3. Install and configure the plugin. In Claude Code, run:
   /plugin install discord@claude-plugins-official
   then:
   /discord:configure <your-token>
4. Launch and pair. Restart Claude Code with:
   claude --channels plugin:discord@claude-plugins-official
   DM your bot on Discord, receive a pairing code, and finalize with:
   /discord:access pair <code>

Once paired, you can treat your Claude Code environment as a remote teammate: send it code snippets, ask it to inspect logs, or request fixes, all from your phone or another machine running Discord or Telegram.

What this unlocks for mobile and remote workflows

For many developers, the biggest shift is practical: mobile AI coding without self-hosted gymnastics. Before Channels, if you wanted to check a build, trigger a job, or ask your AI to patch something while away from your desk, you typically needed a setup like OpenClaw on dedicated hardware.

With Channels:

• Your Claude Code terminal or server acts as the execution environment.
• Telegram or Discord becomes your lightweight control interface.
• You can fire off tasks and get notifications when they’re done, wherever you are.

Anthropic also ships a “Fakechat” demo—a local-only chat UI that simulates the messaging workflow without exposing your environment to the internet. It lets you test how messages are routed and how Claude responds before connecting a real Telegram or Discord bot. That aligns with Anthropic’s stated preference for a cautious, “research preview” rollout where developers understand the event flow and risk surface before going live.

MCP and the open–closed balance

Claude Code itself remains proprietary and tied to Anthropic’s paid tiers (Pro, Max, Enterprise). But the company is leaning hard on the open Model Context Protocol as the backbone for Channels.

That has a few important implications for developers:

• Open connectors on top of a closed core. While the Claude “brain” is closed, the Telegram and Discord plugins are hosted in official Anthropic GitHub repositories. That likely enables community contributions, customization, or forks, even though the underlying model remains commercial.
• Extensibility beyond Telegram and Discord. Because MCP is the unifying layer, the community isn’t limited to whichever channels Anthropic ships. In principle, developers can build their own MCP-based connectors for Slack, WhatsApp, or other services, reusing the same pattern as the official plugins.
• Alternative to fragmented open-source agents. Many open-source agent frameworks are powerful but piecemeal, requiring you to glue together tools, runtimes, and messaging adapters. Anthropic’s strategy is to offer a coherent, supported stack while still inviting open-source work at the integration layer.

This “proprietary engine on open tracks” approach is becoming a recurring pattern across the AI industry, and Channels is a concrete example of how it plays out in developer tooling.

Developer and community response

Early reactions from AI observers and developers on X framed Channels explicitly in terms of the OpenClaw comparison. Multiple users described Anthropic’s pace—adding texting-style interfaces, support for thousands of MCP tools, and autonomous bug-fixing in a short timeframe—as unusually fast for a major AI provider.

Some power users who previously invested in 24/7 OpenClaw setups now see less need for dedicated hardware. One common sentiment is that with Channels, you no longer have to buy and maintain a separate Mac mini just to keep your agent online; instead, persistence is handled by Claude Code itself running in a background environment.

AI creator Matthew Berman characterized the shift by saying Anthropic has essentially “built OpenClaw” into Claude Code. That captures the consensus among early adopters: Anthropic has absorbed the most attractive aspects of the open-source agent wave—multi‑channel messaging and long-lived sessions—while maintaining the reliability and polish expected from a tier-one AI platform.

At the same time, the choice is no longer binary. Developers can now weigh:

• a fully local, open-source agent with maximum control but higher responsibility for security and uptime
• or a managed, branded agent harnessed through Claude Code, with less operational burden but within Anthropic’s commercial and policy boundaries

With Claude Code Channels, Anthropic has brought that tradeoff into much sharper focus—and, for many, made an always-on AI coding companion as simple as adding a bot to Telegram or Discord.