Code2Onto: Multi-Agent System for Code-Driven Ontology Population

Ontologies have become essential for knowledge representation, enabling structured, machine-readable data encoding to support interoperability, reasoning, and reuse across various domains. However, the adoption of ontologies in real-world workflows remains challenging due to the complexity of ontology population, which often requires manual intervention, domain expertise, and customized pipelines. Existing AI-driven approaches have focused primarily on unstructured or semi-structured data, leaving the unique challenges of program code and its runtime context largely unaddressed. This paper introduces Code2Onto, a novel framework that simplifies the transformation of source code, including runtime context, into validated ontology instances with minimal developer effort. Code2Onto leverages Large Language Models as orchestrators for automating source code analysis, runtime context extraction, and ontology population, supported by a modular Multi-Agent System built on the Model Context Protocol. Besides a detailed description of architectural components, we evaluate the framework's performance in different demonstration tasks, discuss its limitations, and elaborate on future research directions.