A Telegram MCP server lets an AI client send and read Telegram messages — reply, react, edit, or push notifications from your agent. The simplest path uses a bot token; a heavier path uses your account via MTProto. Here is the setup.
Step 1 — create a bot
In Telegram, open @BotFather, send /newbot, and copy the token it returns (including the leading number and colon, e.g. 123456789:AA...). Message your new bot once so it can talk to you.
Step 2 — configure the client
A representative config (package/runtime vary by implementation — Node, Python/Telethon, or Bun):
{
"mcpServers": {
"telegram": {
"command": "npx",
"args": ["-y", "mcp-telegram"],
"env": {
"TELEGRAM_BOT_TOKEN": "123456789:AA...",
"AUTHORIZED_CHAT_ID": "your_chat_id"
}
}
}
}
Restart the client; see MCP config file location. For Claude Code, there is also an official Telegram plugin you can enable directly.
Bot vs account (MTProto)
- Bot token — easiest and safest; the bot only sees chats it's in. Good for notifications and bot-channel automation.
- MTProto / account servers (e.g. Telethon-based) — act as your account: read all your chats, manage groups. Far more powerful and far more sensitive — only run a server you trust.
Verify
Ask: "Send me a Telegram message: build finished ✅." If it arrives, you're connected.
Common problems
- Bot can't message you — you haven't started a chat with the bot, or the chat ID is wrong.
unauthorized— wrong token; regenerate via BotFather.- No tools in client — see Cursor MCP not working.
Security
A bot token controls your bot; an account/MTProto server controls your whole Telegram. Keep tokens out of shared config and review account-level servers carefully — see how to vet an MCP server.
Going further
Telegram pairs with monitoring and CI servers for push alerts. Browse the communication category or curated loadouts.