Ava: You're explaining a system to an AI assistant, and instead of text you get a diagram back. That's the idea behind uml-mcp.

Andrew: Wait, so I describe it and it draws it?

Ava: That's it. uml-mcp is an open-source Model Context Protocol server, an M C P server, that generates Mermaid, PlantUML and Kroki diagrams. You can give it a natural-language prompt, or write the diagram source yourself.

Andrew: Where do I use it?

Ava: In Cursor, through its M C P tools. Or in ChatGPT through a custom GPT.

Andrew: What kinds of diagrams?

Ava: The main tool is called generate underscore U M L, plus related ones, and it covers sequence, class, flow and deployment sketches.

Andrew: And why three engines?

Ava: Multiple render backends, so you pick the one that suits your diagram. The server abstracts them.

Andrew: Does Antoine actually use it?

Ava: Yes. It's used across this portfolio blog for inline diagrams, with Mermaid source files and S V G output.

Andrew: Where does it come from?

Ava: It follows two earlier projects: a ChatGPT plugin path, and a Python PlantUML client. There's also a hosted M C P endpoint and a documentation site.

Andrew: And it's open source.

Ava: MIT licensed. The README covers Python dependencies and the Cursor setup, and there's a blog post with the full story.