MCP Integration example with Telegram

A Python-based Model Context Protocol (MCP) client and server that enables LLMs to send notifications via Telegram and receive user responses.

Features

• Send text alerts to a Telegram chat with customizable urgency levels
• Check for user replies with configurable timeouts
• Flexible template for integration with MCP-compatible LLM applications

MCP vs Function Calls or Agent tools

MCP Protocol (Model Context Protocol):

• Separates the LLM’s reasoning (the “context”) from the execution of external actions. For example, the LLM decides what action to take and the MCP server handles how to execute that action.
• This decoupling means you can update or swap out external tools without changing the core LLM model or its prompts.
• It also allows for safer and more controlled execution of actions.

LLM Function Call or Agent:

• Direct function calls or agent-based methods often mix decision-making and execution logic, which can make the system less modular.
• Changes in how an action is performed may require reengineering the agent or modifying the LLM’s prompts, making maintenance harder.

Prerequisites

• Python 3.11+
• A Telegram bot token (from @BotFather)
• Your Telegram chat ID

Installation

1. Clone the repository:

   git git@github.com:arananet/mcp-telegram-poc.git
   cd mcp-telegram-poc
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Create the virtual environment:

   python3 -m venv .venv
   

4. Copy .env.example to .env and fill in your Telegram credentials:

   cp .env.example .env
   nano .env
   

Configuration

Edit .env with:

• TELEGRAM_BOT_TOKEN=your_bot_token_here
• TELEGRAM_CHAT_ID=your_chat_id_here
• OPENAI_API_KEY=your_openai_apikey

Getting a Telegram Bot Token

1. Chat with @BotFather on Telegram.
2. Send /newbot and provide a name (e.g. mcp_example) and then provide an user name (e.g. mcp_example_bot) after this telegram botfather will reply with the token.
3. Copy the token provided.

Finding Your Chat ID

1. Send a message to your bot (go to the created bot and send some example, e.g. hi).
2. Visit https://api.telegram.org/bot<YOUR_BOT_TOKEN>/getUpdates.
3. Find the chat object's id field. You will see something like this (e.g. "chat":{"id":107184517,"first_name":"Edu","last_name":"Arana"...) the Chat ID is 107184517.

Usage

Running Standalone

python3 src/mcp_handler.py

Programmatic Usage

For testing sending and receiving an alert:

python3 examples/test_client.py

Agent integration

For testing the integration of the MCP using an agent with autogen 0.4 framework

python3 examples/test_autogen_agent.py

Testing Locally

Use the test script (Linux/macOS):

chmod +x test_server.sh
./test_server.sh

Available Tools

send_alert

• Parameters:
  • message (str, required): Message to send (Markdown supported).
  • project (str, required): Project name.
  • urgency (str, optional): "low", "medium", "high" (default: "medium").
• Description: Sends a Telegram alert and waits for a reply.

check_reply

• Parameters:
  • message_id (int, required): Message ID to check.
  • timeout_seconds (int, optional): Wait time (default: 60).
• Description: Polls for replies to a specific message.

Development

• Install dependencies: pip install -r requirements.txt
• Run server: python3 src/mcp_handler.py
• Test client: python3 examples/test_client.py
• Test Agent: python3 examples/test_autogen_agent.py

Examples

The examples folder contains:

• test_server.py: Runs the server standalone.

• test_client.py: Demonstrates client interaction.

• test_autogen_agent.py: Demostrates the MCP integration between an AutoGen Agent

• In the images folder you can find some screenshots of the testings.

Developer

• Eduardo Arana