This session was a really fun one to record. Together with my friend, Maciek Jarka, we walked through the full process of creating a custom MCP server and integrating it with Microsoft Copilot Studio. We started with the basics of setting up the Python environment and then moved into building the actual server that powers the travel assistant. From there we explored the steps needed to host the server on a public endpoint, connect it inside Copilot Studio, and work through the inevitable debugging moments that happen when you are dealing with connectivity and authorization.

The example we used in the video is a travel booking assistant that can help travel agencies or corporate travel teams with day to day tasks. The interesting part is how easily this same pattern can be reused across so many other business scenarios. Any system that you want Copilot Studio to talk to can be wrapped with an MCP server, which makes this approach incredibly flexible.

The video also includes a collection of helpful resources such as the MCP server code sample, FastMCP documentation, security guidance, the OutScraper API we used for hotel reviews, and a community explanation of why FastMCP is the recommended SDK for Copilot Studio integrations.

It is a practical, hands on session that blends real world architecture with experimentation, and it opens the door for a lot of creative AI agent ideas across different industries.