Not too long ago, I started building my first Microsoft Copilot Cowork plugin and found that the system I wanted Cowork to connect to needed an MCP server on top of it. This series is what I learned going down that rabbit hole, using one running example: a library with an SQL database of books.
The series
- MCP, Part 1: What it is and how it works
What MCP is, why AI apps need MCP servers, how hosts, clients and servers fit together, and when to build your own. - MCP, Part 2: How to design your MCP server
How to shape an MCP server’s tools, resources and prompts, the anti-patterns to avoid, and C# examples for the library. - MCP, Part 3: Build and test your MCP server
Building the library MCP server in C# on Azure SQL, and testing it locally with the MCP Inspector and AI apps in VS Code. - MCP, Part 4: Deploy your MCP server to Azure and add sign-in
Hosting the library server in Azure and replacing the test sign-in with Microsoft Entra ID. - MCP, Part 5: Connect your MCP server to Copilot Cowork
Packaging the library server as a Copilot Cowork plugin with skills, setting up Entra ID sign-in, and testing it in Cowork. - MCP, Part 6: Connect your MCP server to Claude (coming soon)
Connecting the library server to Claude with a plugin. - MCP, Part 7: Connect your MCP server to ChatGPT (coming soon)
Connecting the library server to ChatGPT with a plugin, which also works in Codex.
Get new posts by email
New posts on AI, agents and Power Platform, straight to your inbox. No spam, unsubscribe anytime.