Telegram AI Assistant Bot

What is Telegram AI Assistant Bot?

Telegram AI Assistant Bot is a reasoning-driven AI agent named Cortex-R that interacts with users via Telegram. It uses multiple Model Context Protocol (MCP) servers to perform tasks such as web searches, document processing, mathematical calculations, and Google services integration (Gmail, Google Sheets). The bot is built for users who want an intelligent assistant accessible through Telegram.

How to use Telegram AI Assistant Bot?

Clone the repository, sync dependencies with uv sync, set environment variables in .env (including Gemini API key and Google credentials), then run uv run telegram_agent.py. On Telegram, send /start to begin, then send queries or tasks like "Find the current F1 standings and share them in a Google Sheet" or "Send an email to [email protected] with the subject 'Meeting Update'".

Key features of Telegram AI Assistant Bot

• Telegram integration for messaging and commands.
• Multi-MCP server support (Gmail, Google Sheets, web search, documents, math).
• Dynamic tool discovery from MCP servers during initialization.
• Customizable agent strategies (conservative, retry-once, explore-all).
• Memory management to store and retrieve past interactions.
• Configurable persona for tone, verbosity, and behavior.

Use cases of Telegram AI Assistant Bot

• Automate email management: read, send, or trash emails via Gmail.
• Create, update, and share Google Sheets from Telegram.
• Perform web searches and fetch page content using DuckDuckGo.
• Process documents for semantic search and indexing.
• Execute mathematical calculations on the fly.

FAQ from Telegram AI Assistant Bot

What runtime dependencies does the bot require?

It requires Python, the uv package manager, and API keys/credentials (Gemini API key, Google service account, OAuth client credentials). Configuration is done via a .env file and YAML/JSON config files.

Which transport protocols do the MCP servers use?

The bot supports both stdio and sse transport protocols to communicate with its MCP servers, as configured in profiles.yaml.

Where are user data and memory stored?

Memory items (tool outputs and user queries) are stored in the bot’s memory system. Google Drive is used for file storage (folder ID set in .env), and Google APIs rely on token and credential files.

Is there a limit to the number of MCP servers I can add?

The README does not specify a hard limit, but the bot dynamically discovers tools from all servers listed in profiles.yaml, and multiple servers (e.g., Gmail, Google Sheets, math, documents, web search) are demonstrated.