DEV Community

Cover image for How I Built an MCP Server to Give AI Assistants Real Telegram Powers
Alexey Leshchenko
Alexey Leshchenko

Posted on

How I Built an MCP Server to Give AI Assistants Real Telegram Powers

#telegram #mcp #ai

I've been working on AI integrations for a while, and one thing always bugged me: why can't AI assistants just... use Telegram like humans do? Search conversations, send messages, manage contacts - without all the complexity.

Building fast-mcp-telegram was my answer. It's a complete MCP server that lets AI assistants interact with Telegram naturally.

The Problem I Needed to Solve

Working with AI assistants, I kept running into limitations. They could analyze data, but actually using Telegram was clunky. Direct API calls required complex session management, bot frameworks were for user-facing chatbots, and search tools couldn't send messages or manage contacts.

I needed something that gave AI assistants full Telegram capabilities in a natural way.

What Makes This Different

After trying various approaches, I built something specifically for AI assistants:

  • MCP Tools: Direct Telegram access through the invoke_mtproto tool and standard messaging/search functions
  • HTTP Bridge: For no-code tools like n8n and Make.com that can't use MCP directly
  • Web Setup: Handles authentication and generates config files automatically
  • Production Support: Bearer tokens, session isolation, and proper error handling
  • Smart Search: Multi-query support with deduplication and filtering for AI assistants
  • Full Messaging: Send, edit, reply, share files, even message phone numbers not in contacts
  • File Handling: Works with URLs or local files, handles security and albums

Where This Fits In

Other Telegram projects solve specific problems:

  • Search tools are great for finding messages, but can't send replies or manage contacts
  • Bot frameworks work for user-facing chatbots, but not for AI assistants needing programmatic access
  • Other MCP servers connect specific tools; this brings the entire Telegram ecosystem to AI assistants via direct MTProto API access

Getting Started

Try the live demo: https://tg-mcp.redevest.ru/setup - log in and download a config file, no installation needed.

For developers:

pip install fast-mcp-telegram
fast-mcp-telegram-setup \
  --api-id="your_api_id" \
  --api-hash="your_api_hash" \
  --phone-number="+123456789"
Enter fullscreen mode Exit fullscreen mode

Full docs: https://github.com/leshchenko1979/fast-mcp-telegram/#readme

What I Use It For

  • Daily news summaries: n8n automation searches my subscribed channels for the last 24 hours and sends AI-summarized digests
  • Smart spam detection: Uses HTTP-MTProto Bridge to get full user profiles (beyond regular Bot API) for better spam scoring
  • Content creation: AI assistants analyze my previous posts' formatting to maintain consistent style
  • Customer service: AI can respond to Telegram inquiries instead of just reading them
  • Research workflows: Search across channels and summarize findings

Wrapping Up

This started as a solution to my own AI-Telegram integration frustrations. The MCP protocol makes it natural for AI assistants to use, and the full feature set enables real applications beyond just demos.

If this sounds useful, try the demo - setup takes 2 minutes.

Demo: https://tg-mcp.redevest.ru/setup
GitHub: https://github.com/leshchenko1979/fast-mcp-telegram
Docs: https://github.com/leshchenko1979/fast-mcp-telegram/blob/master/docs/

This post is AI-assisted. But nowadays, everything is, right?

Top comments (0)