Key points are not available for this paper at this time.
This perspective explores the emerging integration of multimodal large language models (MLLMs), agentic artificial intelligence (AI) capabilities, and real-world healthcare infrastructure. Agentic AI refers to systems that can analyze and interpret complex data and autonomously perform goal-directed tasks, potentially transforming AI from passive tools into active participants in clinical workflows. To support such capabilities, new infrastructure is needed to enable secure and dynamic connectivity with diverse health information systems. Despite early progress, agentic AI faces major challenges, including limited interoperability across healthcare systems, underdeveloped tool ecosystems, and risks in autonomous decision-making. The recently proposed model-context protocol (MCP) aims to meet this need by offering a conceptual framework that could allow AI agents to retrieve, interpret, and act upon clinical data across environments such as electronic health records, imaging systems, and laboratory databases. We examine how MCP-style integration might enhance interoperability, contextual responsiveness, and safety in future medical AI applications. We also outline key limitations, including immature tool ecosystems, legal and ethical uncertainties, and computational constraints. While MCP does not yet constitute a widely adopted standard, it could provide a potential foundation for developing modular, reliable, and secure agentic systems. Looking ahead, the convergence of MLLMs, agentic AI, and MCP-style integration could enable more adaptive and collaborative AI systems in healthcare if key technical, ethical, and regulatory challenges are carefully addressed. This perspective synthesizes these concepts to propose a framework for integrating MLLMs, agentic AI, and MCP, outlining key applications and enhancing clinical interoperability.
Choi et al. (Sat,) studied this question.