| Kurzfassung
|
This thesis investigates how AI coding agents can support model-driven engineering directly within developers’ existing environments. It presents PCM-MCP, a Model Context Protocol server for the Palladio Component Model that exposes modeling operations as tool calls for querying, modifying, and validating model instances. Instead of letting agents write XML model files directly, PCM-MCP performs the modeling operations itself, ensuring syntactic well-formedness by construction while enabling agents to iteratively detect and fix remaining OCL constraint violations. The approach supports interactive AI-assisted modeling workflows not covered by prior standalone validation pipelines. An evaluation with 20 test cases from two Java microservice systems generated 600 PCM instances across three LLMs. All generated instances were syntactically well-formed and free of OCL violations. Compared with expert reference models, they achieved about 65% mean Jaccard similarity under relaxed name matching, with lower strict scores due to naming deviations.
|
