What Is Claude Code Channels? (Complete Explanation)

Claude Code Channels launched on March 20, 2026. If you heard about it today and want to understand what it actually is before deciding whether it is worth setting up, this is the guide for you.

No hype. No step-by-step tutorial. Just a clear explanation of what it is, how it works, and whether it belongs in your workflow.

‍

Key Takeaways

• Launched March 20, 2026: Channels is a brand-new research preview inside Claude Code, not a standalone product or a general Claude feature.
• Phone controls, terminal executes: your messaging app sends instructions; your local Claude Code session does the actual work; nothing runs in the cloud.
• Two-way communication: you send instructions from Telegram or Discord, Claude replies back through the same channel with progress updates, results, and completions.
• Requires Claude Code v2.1.80 plus, Bun runtime, and claude.ai login: Console and API key authentication are not supported in the research preview.
• Session must be open to receive messages: Channels is not always-on; close your terminal and the channel goes dark until you reopen the session.
• Anthropic's answer to OpenClaw: same cross-device messaging concept, no dedicated hardware, no manual server setup, backed by Anthropic's security model out of the box.

‍

AI App Development

Your Business. Powered by AI

We build AI-driven apps that don’t just solve problems—they transform how people experience your product.

Let's talk

‍

‍

The Problem Claude Code Channels Solves

To understand why Channels matters, you need to understand what working with Claude Code looked like before it shipped.

‍

Claude Code is a powerful agentic coding tool. You open your terminal, start a session, give Claude a task, and it works through your codebase autonomously. The problem is everything that happens after you start that task.

You had to stay at your terminal. A long build, a complex refactor, a multi-file debugging session, all of it required you to sit and watch. If you stepped away, you missed permission prompts that paused the session.

‍

If you wanted to add context mid-task, you had to be at your desk to type it. The interaction model was entirely synchronous.

• The ask-and-wait problem: every interaction required your physical presence at the terminal; there was no way to check progress, send an instruction, or get notified of completion from anywhere else.
• No mobile access to your coding agent: your AI coding assistant ran locally and communicated only through the terminal or Claude Desktop; your phone had no direct line to it.
• What OpenClaw was doing that Claude Code was not: OpenClaw became viral in early 2026 precisely because it let developers message an AI agent through Telegram, WhatsApp, Discord, and iMessage; Claude Code had none of this native capability.
• The gap was a real workflow constraint: developers running long agentic tasks either stayed glued to their computer or lost context when they stepped away; neither option was acceptable for serious development work.

Channels is Anthropic's answer to that gap. It ships the same cross-device messaging capability that made OpenClaw compelling, built natively into Claude Code with Anthropic's security model and no external server requirement.

‍

What Claude Code Channels Actually Are

‍

A Channel is an MCP server that pushes events into your running Claude Code session so Claude can react to things that happen while you are not at the terminal.

‍

That one sentence from Anthropic's official documentation is the cleanest definition available. Everything else is context around it.

• The core idea is bidirectional messaging into a live session: you send a message from your phone through Telegram or Discord, the Channel forwards it into your active Claude Code session, Claude processes it, does the work, and replies back through the same channel.
• It is two-way: this is not a notification-only system; you can send new instructions mid-task, receive progress updates, get notified when something completes or fails, and steer the agent from your phone.
• The shift is from synchronous to asynchronous: instead of sitting at your terminal waiting and watching, you set a task running, step away, and interact with it on your own schedule from wherever you are.

Think of it as a messaging bridge between your phone and your terminal. Your phone becomes a remote control for a Claude Code session already running on your machine.

‍

What You Can Actually Do With Claude Code Channels (Real Use Cases)

The practical value of Channels is clearest in specific developer scenarios that previously required staying at a desk.

• Start a long build or task, step away, get notified when it is done: kick off a complex refactor or a long test suite run, leave your desk, and receive the completion message directly in Telegram without polling your terminal manually.
• Send new instructions to Claude mid-task from your phone: realize mid-commute that you forgot to include a specific constraint in your instructions; send it via Telegram and Claude incorporates it into the running session without requiring you to return to your desk first.
• Get alerts when a bug is fixed, a test fails, or a build completes: Claude can react to CI results, test outputs, and build events pushed through the channel, effectively making your phone an event monitor for your development environment.
• Lightweight incident response without being at your desk: triage an alert, check a build status, or instruct Claude to investigate a failing test from wherever you are without opening your laptop in an inconvenient location.
• Code review on the go: send a review instruction from your phone, have Claude analyze the relevant files and summarize the output, and read the response in Telegram before you get back to your desk.

‍

How Claude Code Channels Work

The architecture is straightforward once you understand MCP as the connective layer.

‍

MCP, the Model Context Protocol that Anthropic open-sourced and donated to the Linux Foundation in December 2025, is the standard that makes Channels possible. Think of it as a universal connector that lets external tools communicate with Claude in a standardized way.

‍

• A Channel is an MCP server that sits between your messaging app and Claude Code: when you install the Telegram or Discord plugin and start Claude Code with the --channels flag, you are spinning up a polling service that bridges your messaging platform and your active session.
• When you send a message, the MCP server forwards it into your active Claude Code session: the server logs into Telegram as a bot, receives your DM, and injects that message directly into the running Claude Code session as an event.
• Claude processes it, does the work, and replies back through the same channel: the three tools the Telegram plugin exposes to Claude are reply, react, and edit message; Claude uses these to communicate back to your phone.
• Events only arrive while the session is open: this is not a cloud service and not a background daemon; if your terminal session closes, messages stop arriving; for always-on setups you run Claude in a persistent background process.

‍

What Platforms Claude Code Channels Support Right Now

Channels launched with official Anthropic-maintained plugins for the following platforms.

• Telegram: the most fully documented plugin with DM support, group support, photo delivery, typing indicators, and three MCP tools for reply, react, and message editing.
• Discord: works through a bot you create in the Discord Developer Portal; supports DMs and server channels with similar two-way messaging capability.
• iMessage: listed as a supported platform in Anthropic's documentation alongside Telegram and Discord.
• Webhooks: for custom integrations where you want to connect Claude Code to a system that does not have a pre-built plugin; allows developers to build their own channel for any platform.
• Fakechat: an officially supported demo channel that runs a chat UI on localhost with nothing to authenticate; the recommended starting point before setting up a real platform, useful for testing the concept before committing to Telegram or Discord setup.

Because Channels is built on MCP, community members can build connectors for Slack, WhatsApp, and other platforms without waiting for Anthropic to ship them officially.

‍

What Claude Code Channels are NOT (Clear the Confusion)

Several things Channels are not worth stating explicitly because the framing around cross-device AI can create wrong expectations.

• Not a cloud service: Claude Code still runs entirely on your machine; Channels is a messaging bridge to your local session, not a hosted AI service that runs independently in the cloud.
• Not always-on by default: your session must be open to receive messages; close the terminal, close Claude Desktop, and the channel goes dark; always-on requires running Claude in a persistent background process yourself.
• Not a replacement for the Claude mobile app: the Claude mobile app is a general-purpose interface for chatting with Claude; Channels is specifically for controlling your Claude Code session from a messaging platform.
• Not available via API key or Console auth: Channels requires a claude.ai login; Console and API key authentication is not supported in the research preview.
• Not a general chatbot bridge: you are not creating a Claude chatbot in Telegram; you are creating a remote control interface for your specific Claude Code session and the codebase it has access to.
• Not GA yet: Channels is a research preview as of March 20, 2026; the protocol and plugin commands may change based on feedback before it reaches general availability.

‍

Who It Is For (And Who It Is Not)

• Developers running long agentic tasks who do not want to babysit the terminal: the clearest beneficiary is any developer who regularly starts tasks in Claude Code that take more than a few minutes and previously had no way to interact with them remotely.
• Teams who want mobile access to their AI coding agent: engineering teams doing on-call work, incident response, or remote troubleshooting benefit from being able to triage and instruct Claude Code from their phones without requiring a laptop.
• Engineers doing remote troubleshooting or on-call work: sending a Telegram message to your Claude Code session asking it to check a failing endpoint is a meaningfully better experience than opening your laptop at 11pm to type into a terminal.
• Not for casual Claude users: Channels is specifically a Claude Code feature; it is not relevant for users who primarily use Claude for chat, writing, or general productivity tasks.
• Not for API or Console users: the research preview requires a claude.ai subscription on Pro, Max, Team, or Enterprise plans; API key and Console authentication are not supported.

‍

How Claude Code Channels Compares to What Came Before

vs Remote Control (the old method):

Claude Code previously had a Remote Control feature that let you connect your phone to a running session; early users described it as flaky and inconsistent;

Channels is purpose-built for this use case, more stable, and built on the standardized MCP protocol rather than a proprietary connection.

‍

vs OpenClaw:

OpenClaw achieved the same cross-device messaging concept through a self-hosted architecture requiring Node.js 22 plus, a WebSocket gateway, a dedicated server or Mac Mini, and extensive configuration.

Channels achieves the same core outcome with a plugin install and the --channels flag, no dedicated hardware, and Anthropic-backed security rather than community-maintained safety measures.

‍

What Claude Code Channels can and OpenClaw not:

Anthropic's native security model means no third-party server handling your credentials; official repository maintenance means no dependency on community contributors staying engaged; and the MCP foundation means community-built connectors can extend the platform without forking the core tool.

‍

Current Limitations of Claude Code Channels (Before You Get Excited)

Channels is genuinely useful. It is also a research preview with specific constraints that matter for how you use it.

• Research preview, not production-ready: the protocol and plugin commands may change before GA; do not build critical workflows that depend on exact command syntax staying stable.
• Session must stay open: messages do not queue when your session is closed; if Claude Code terminates mid-task, messages sent while it was offline are gone; for reliable remote interaction you need a persistent terminal or background process.
• Permission prompts pause the session: if Claude hits a permission prompt while you are away, the session pauses and waits for local approval; you cannot approve permissions remotely; for fully unattended use you can use --dangerously-skip-permissions only in environments you fully trust.
• No message history for the Telegram bot: Telegram's Bot API does not expose message history; the bot only sees messages as they arrive after the session starts; there is no way to fetch earlier messages you missed.
• Requires Bun runtime, claude.ai login, and Claude Code v2.1.80 plus: the pre-built channel plugins are Bun scripts; Node.js will not work; verify your Claude Code version before trying to set up Channels.

‍

What Is Likely Coming Next (Our Expectations)

The limitations of the current research preview point directly at what the next versions will address.

• Always-on cloud-hosted sessions: the "desktop must be on" constraint is the most obvious gap; removing it requires server-side session persistence that Anthropic has not shipped yet but that the product direction clearly points toward.
• More platform support beyond Telegram and Discord: because Channels is built on MCP, community connectors for Slack, WhatsApp, and other platforms are already technically possible; official support for more platforms is likely as the research preview matures.
• Better remote permission handling: the current requirement to approve permissions locally is the second-biggest workflow gap; handling permission prompts remotely through the messaging interface is the natural next improvement.
• Moving from research preview to GA: the protocol and command syntax will stabilize before general availability; teams building workflows around Channels should track the official documentation for breaking changes during the preview period.

‍

AI App Development

Your Business. Powered by AI

We build AI-driven apps that don’t just solve problems—they transform how people experience your product.

Let's talk

‍

‍

Want to Build AI-Powered Development Systems for Your Team?

Understanding Claude Code Channels is the first step. Building AI-powered development workflows and systems that actually fit your team's operations requires a different kind of thinking.

At LowCode Agency, we are a strategic product team that designs, builds, and evolves custom AI-powered tools, automation systems, and business software for growing SMBs and startups. We are not a dev shop.

• AI strategy before tooling: not sure how Claude Code, Channels, and other AI development tools fit into your engineering workflow? Our AI consulting service maps your current setup, identifies where AI creates real leverage, and defines a build plan before any development begins.
• Custom AI agent development: we design and build AI agents built around your specific workflows rather than configuring general platforms that approximate what you need.
• Production-grade reliability: every system we build handles real operational load with proper error handling and output quality your team depends on every day.
• Full product team on every project: strategy, UX, development, and QA working together from discovery through deployment and beyond.
• Long-term partnership after launch: we stay involved after delivery, evolving your AI systems as your operations and requirements grow.

We have shipped 350+ products across 20+ industries. Clients include Medtronic, American Express, Coca-Cola, and Zapier.

If you are serious about building AI-powered systems that work reliably at production scale, let's talk at lowcode.agency/contact.