End-to-End AI Agent Architecture with MCP and TelegramEnd-to-End AI Agent Architecture with MCP and Telegram
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Saif Eddine's avatar
Saif Eddine Kthiri
pro
• Oct 7
Over the past few days, I built an end-to-end AI agent on top of an existing system.
The interesting part was not connecting an LLM to a chatbot.
The real work was in the architecture:
- keeping business logic inside the application
- exposing system capabilities through an MCP server
- keeping the AI agent focused on understanding intent and orchestrating tools
- separating communication channels from the agent, so Telegram works today and WhatsApp or internal chat can be added later without changing the core logic
- preserving existing authorization and application rules instead of duplicating them in the AI layer
The current flow is:
User → Telegram → Channel Layer → AI Agent → Responses API → MCP → Application Services / Actions
The agent can now read application data, understand natural-language requests, execute real operations, maintain conversational context, and return the result through Telegram.
One lesson from this work: The hard part of Agentic AI is not the model itself.
It is designing clear boundaries between reasoning, tools, business rules, permissions, and channels. Once those boundaries are clear, the system becomes much easier to extend, test, and maintain.
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started