Utilizing Model Context Protocol And Ai Models To Enhance Organizations Digital Twin Infrastructures

Sammanfattning

The rapid digital transformation of modern organizations has increased the need to integrate physical operational environments with data-driven Artificial Intelligence (AI) systems. Digital twin platforms provide structured virtual representations of physical assets and their relationships, enabling contex- tualized access to data from buildings, infrastructure and technical systems. Despite their potential, integrating AI systems with digital twins introduces challenges related to secure data access, controlled reasoning, interoperability and governance. This thesis investigates how AI-driven systems can be integrated into digital twin environments in a secure and controlled manner using the Model Context Protocol (MCP). MCP defines a tool-based interaction model that separates AI reasoning from underlying system implementations, allowing con- trolled exposure of external data sources. The study focuses on architectural integration, access control through MCP and the use of digital twin structures as a reference layer for external databases. A design-oriented research methodology was applied through the development of a proof-of-concept implementation. The implementation includes an MCP server enabling controlled interactions be- tween an AI agent, a digital twin platform and external data sources. The solution was evaluated using qualitative industry-driven use cases together with quantitative testing of access-control and data re- trieval mechanisms. The results show that the MCP-based architecture successfully enabled the AI agent to perform multi- step data retrieval across digital twin data and external databases through predefined tools. The digi- tal twin functioned as a reference layer by providing building identifiers, geographic relationships and contextual structures that were used to retrieve and link external energy-consumption data. The proof- of-concept implementation used an Open Authorization (OAuth)-based authentication layer to restrict tool access to users within the organization. In addition, a separate authorization test setup was used to evaluate the feasibility of more granular role-based access control within an MCP server. This test- ing showed that the simulated role-based policy correctly approved or denied all 500 evaluated access requests according to the predefined access policies. The analysis indicates that MCP can provide a structured and controllable approach for integrating AI systems with digital twin environments. However, the results also show that the usefulness of such integrations depends on the quality and completeness of external data sources, as several buildings lacked energy-consumption records or contained inconsistent data values. The thesis concludes that MCP, combined with digital twin identifiers and authorization mechanisms, can support secure and contextualized AI interaction with operational data, while further research is needed on scalability, production deployment, advanced security risks and broader enterprise integration.