Frank Wübbeling (Münster): MCP, API, or Skill? Connecting AI Assistants to University Systems in Practice
Wednesday, 18.11.2026 14:15 im Raum M5
AI assistants such as Claude or ChatGPT can do more than write text. They
can also call tools, for example to query databases, edit web pages or book
rooms. The Model Context Protocol (MCP) is the standard for connecting such
tools. This is a hands-on talk. We look at how an MCP server is built,
starting with the smallest possible server, and how MCP differs from two
related ideas: a conventional API and a skill (packaged instructions for the
assistant). The examples are the Math*DB server, which makes the MaRDI
knowledge graphs MathModDB and MathAlgoDB usable by language models (joint
work with Björn Schembera), and the connection to Imperia, the university's
content management system. Imperia has no official interface, so we first
built an API for common editing tasks from the behaviour of its web
interface and then put an MCP server on top of it. The talk ends with an
overview of further connections (Indico, Learnweb, room booking, Mattermost,
...) and what works well and what doesn't.
Angelegt am 02.10.2026 von Stephan Rave
Geändert am 16.11.2026 von Stephan Rave
[Edit | Vorlage]