Claude Code Channels: Telegram and Discord in 10 minutes

What Channels does

Claude Code Channels is an MCP (Model Context Protocol) plugin that connects Claude Code to Telegram and Discord. You message your bot with a task, and Claude Code receives it as if you had typed it directly into the terminal. The agent executes the task in your current project directory with full access to your tools, files, and shell.

The practical use case: you are away from your computer but want to kick off a deployment, run a database migration, or ask Claude to investigate a production error. You send the task from your phone, and it runs against your real project context.

Prerequisites

• Claude Code installed and working. If you have not set this up yet, follow the official Claude Code docs.
• An Anthropic subscription (Pro or Max). Claude Code is not available on the free tier.
• A Telegram or Discord account for the messaging side.

Step 1: Install the Channels MCP plugin

From inside a Claude Code session, run:

claude mcp install channels

This registers the Channels plugin with your Claude Code installation. The plugin activates automatically on subsequent sessions. Verify it is installed:

claude mcp list

You should see channels in the output with status active.

Step 2: Set up a Telegram bot

1. Open Telegram and message @BotFather.
2. Send /newbot. Choose a name (e.g., "My Claude Agent") and a username (e.g., my_claude_agent_bot).
3. Copy the API token BotFather gives you.
4. Configure the Channels plugin:

claude mcp configure channels \
  --set telegram.token="7123456789:AAF..." \
  --set telegram.allowed_users="YOUR_USER_ID"

Replace YOUR_USER_ID with your numeric Telegram ID. Get it by messaging @userinfobot.

Step 3: Set up Discord (optional)

If you prefer Discord or want both channels active:

1. Go to the Discord Developer Portal and create a new application.
2. Under the Bot tab, click "Add Bot" and copy the bot token.
3. Under OAuth2, generate an invite URL with the bot scope and Send Messages + Read Message History permissions.
4. Invite the bot to your server using the generated URL.
5. Configure:

claude mcp configure channels \
  --set discord.token="MTIz..." \
  --set discord.allowed_users="YOUR_DISCORD_USER_ID"

Step 4: Send your first task

Start (or restart) Claude Code in your project directory so the Channels plugin activates:

claude

Open your Telegram bot conversation and send a message:

List all TODO comments in the codebase and summarize what each one is about

Claude Code receives the message, processes it against your project files, and sends the result back to your Telegram chat. The response time depends on your subscription tier and the task complexity. Simple queries return in 10 to 30 seconds. Complex multi-file operations can take a few minutes.

Limitations to know about

Channels is useful but has real constraints you should understand before relying on it:

• Session-bound. If Claude Code is not running on your machine, tasks queue but do not execute. There is no background daemon. You must have an active terminal session for the agent to work.
• Single provider. Channels only works with Claude Code and Anthropic models. You cannot swap in OpenAI or a local model. The LLM is whatever Claude Code is configured to use.
• No persistent memory between sessions. If you restart Claude Code, the agent starts fresh. Conversation history from Channels does not carry over unless you use Claude Code's built-in memory features.
• Subscription limits apply. Every Channels task consumes tokens from your Claude subscription. Heavy use from Telegram can hit rate limits the same way heavy terminal use does.

When to use Channels vs. a standalone agent

Channels is the right choice when you already use Claude Code daily and want mobile access to the same agent. It adds zero operational overhead because it runs inside your existing session.

A standalone agent framework like OpenClaw or Hermes is better when you need always-on operation, multiple LLM providers, or agents that run independently of your laptop being open. See the agentic AI overview for the full comparison.

Frequently asked questions