Agentic SPARQL: Assessing Intelligent Agents Utilizing SPARQL-MCP on the Federated KGQA Benchmark

Recent advancements in AI have seen the emergence of 'agentic' applications, particularly those leveraging the Model Context Protocol (MCP) to enable large language models (LLMs) to interact with various tools. This integration allows for more sophisticated problem-solving capabilities by tapping into external data sources. SPARQL endpoints, which are publicly accessible, serve as an ideal medium for merging diverse datasets through the MCP. This is achieved by adhering to a standardized protocol and query language, utilizing consistent metadata formats, and enabling the federation of queries. In this study, we delve into the effectiveness of SPARQL-MCP-driven intelligent agents in executing federated SPARQL queries. We first outline the process of adapting an existing Knowledge Graph Question Answering (KGQA) benchmark to incorporate agentic federated KGQA (FKGQA). Subsequently, we implement and assess the integration of SPARQL federation with LLM agents via MCP, focusing on aspects such as endpoint discovery, source selection, schema exploration, and query formulation. Our findings provide valuable insights into the architectural alternatives and enhance the existing methodologies for automated SPARQL query federation in conjunction with agentic AI.