Telegram MCP

MCP server that lets AI agents communicate with users via Telegram.

Installation

pip install telegram-mcp

Configuration

Add to your MCP client config (e.g., Antigravity, Claude Desktop):

{
  "telegram": {
    "command": "telegram-mcp",
    "env": {
      "TELEGRAM_CHAT_ID": "your-chat-id"
    }
  }
}

Get your chat ID: Message @agent_comm_mcp_bot on Telegram and send /start.

That's it! No bot token needed - the client connects to our hosted server.

Tools

interact

Send a message or media to the user via Telegram.

Required Parameters:

• project_name: Project/repository name (e.g., "my-project") - shown in message header
• session_name: Current task/session context (e.g., "Fix login bug") - shown in message header

Optional Parameters:

• message: Text to send (optional if sending media)
• wait_for_reply: Block until user replies (default: False)
• choices: Button labels for quick replies
• media_type: "photo", "video", "audio", or "document" (auto-detected from extension if not provided)
• media_url: URL of remote media to send
• media_path: Local file path to upload and send

Message Format

Messages appear with a header showing context:

🤖 my-project • 📋 Fix login bug
────────────────────
Your message content here...

⏳ Waiting for your reply...

Examples

# Simple status update
interact("my-project", "Deploy", "Working on it...")

# Wait for reply and acknowledge
result = interact("my-project", "Code review", "What's next?", wait_for_reply=True)
interact("my-project", "Code review", f"Got it! Working on: {result['reply']}")

# Yes/No buttons with acknowledgement
result = interact("my-project", "Cleanup", "Delete?", choices=["Yes", "No"], wait_for_reply=True)
if result["reply"] == "Yes":
    interact("my-project", "Cleanup", "Deleting now...")

# Send a photo from URL
interact("my-project", "Results", "Here's the result:", media_type="photo", media_url="https://...")

# Send a local file
interact("my-project", "Export", "Generated image:", media_type="photo", media_path="/path/to/image.png")

# IMPORTANT: Always end with wait_for_reply to keep the loop alive!
interact("my-project", "Complete", "Done! What's next?", wait_for_reply=True)

Agent Guidelines

1. USE TELEGRAM FOR EVERYTHING - All status updates, questions, results
2. ALWAYS END WITH wait_for_reply=True - Never just end; keep the conversation alive
3. ACKNOWLEDGE IMMEDIATELY - After receiving a reply, send acknowledgement before doing work
4. USE MEANINGFUL HEADERS - project_name = repo name, session_name = current task

User Reply Requirement: When responding to wait_for_reply messages, users must swipe the message in Telegram to reply. This ensures replies go to the correct request, especially when multiple agents are active.

Response format

# Text reply
{"reply": "user's text message"}

# Media reply (when user sends photo/video/audio/document)
{
    "reply": {
        "text": "caption if any",
        "media": {
            "type": "photo",
            "file_id": "telegram_file_id",
            "url": "https://api.telegram.org/file/..."
        }
    }
}

# No reply (when wait_for_reply=False)
{"reply": None, "message_id": 123}

get_messages

Get recent conversation history.

result = get_messages(limit=10)

Self-Hosting

If you prefer to run your own server:

{
  "telegram": {
    "command": "telegram-mcp",
    "env": {
      "TELEGRAM_SERVER_URL": "https://your-server.com",
      "TELEGRAM_CHAT_ID": "your-chat-id"
    }
  }
}

See the GitHub repo for server setup.

License

MIT