Poster presented at the CCLS 2026, Potsdam The "DraCor Open Knowledge Graph," developed in the DraCorOS project (OSCARS Cascading Grant), links dramatic texts, their metadata, computational tools, publications, and training materials as semantically enriched Linked Open Data, modelled with CIDOC-CRM, LRMoo, CRMdig, and a slim DraCor extension. Yet querying it requires SPARQL—a barrier that leaves such expert systems under-used by the humanities scholars who would benefit most. This poster presents an approach to democratizing access by implementing a Model Context Protocol (MCP) server that lets Large Language Models autonomously query the graph: users ask in plain language, the model writes and runs the SPARQL, and answers in prose. The central challenge is "docstring engineering"—designing tool documentation that teaches the model the data model, query patterns, namespaces, named graphs, and dataset quirks.
Ingo Börner (Thu,) studied this question.