Evaluating Ontology-Based Function Definitions for MCP Invocation Accuracy in LLM Agent-Based HPC Systems
Yejin Kwon1, Jeongcheol Lee1, Youngbom Park2,*
1 Analysis Platform Team, Department of Supercomputing Acceleration Research, Division of National Supercomputing, Korea Institute of Science and Technology Information, Daejeon, Republic of Korea
2 Artificial Intelligence Information Architecture Lab, Department of Software Engineering, Dankook University, Yongin-si, Republic of Korea
* Corresponding Author: Youngbom Park. Email: 
Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.080249
Received 10 February 2026; Accepted 17 April 2026; Published online 11 May 2026
Abstract
The web-based High-Performance Computing (HPC) platform provides a simulation environment that enables users to perform computational science and engineering tasks through web services, thereby eliminating the need for complex terminal-based environments. Notwithstanding the aforementioned advantages, extant platforms frequently necessitate a considerable degree of user expertise, whilst the intricacy of simulation configuration and execution engenders limitations in terms of accessibility and usability. Furthermore, while Retrieval-Augmented Generation (RAG)-based systems are effective for information retrieval, they are insufficient for accurately constructing and invoking executable service tools. In order to address these limitations, this study proposes a user agent system integrated within a web-based HPC simulation environment, said system being based on an LLM. The proposed system enhances user understanding of available applications and execution workflows, and supports precise configuration and execution of simulations. In order to facilitate practical service tool invocation, the system integrates Model Context Protocol (MCP)-based service tools and introduces an ontology-driven approach for object normalization and relational definition. The system leverages the structured relationships among service tools, transforming LLM outputs into actionable and accurate inputs for service execution. The experimental results demonstrate that the proposed approach significantly improves the accuracy of MCP-based service tool invocation and the appropriateness of responses when compared to conventional RAG-based methods. The proposed system enhances the accessibility and usability of HPC platforms and provides a practical framework for LLM-driven service automation.
Keywords
HPC platform; ontology; RAG (retrieval-augmented generation); LLM (large language model); model context protocol (MCP)
Open Access
Login
Register
Submit a Paper
Propose a Special lssue
Download PDF
Downloads
Citation Tools
