The rapid proliferation of the Internet of Things (IoT) systems has resulted in large volumes of heterogeneous data that are often difficult to access and exploit due to limited interoperability and complex application programming interfaces. Data spaces address these challenges by providing governed environments for secure and semantically interoperable data sharing, commonly relying on standardized interfaces such as the ETSI NGSI-LD API. While powerful, these interfaces are primarily designed for machine-to-machine interaction and remain difficult to use directly by human operators. In this paper, we propose an architecture that enables natural-language access to IoT data stored in a data space by integrating Large Language Models (LLMs) with the Model Context Protocol (MCP). Experimental results using fastMCP and OpenAI API to access a FIWARE-based data space demonstrate that our solution offers accuracy even for prompts that require advanced reasoning.
Athanasopoulou et al. (Thu,) studied this question.