The distributed architectures and intelligent agent integration

Evolution, challenges, and current state of MAS

Abstract

The recent advancement of Artificial Intelligence (AI), particularly generative models, has expanded the applications of intelligent systems in distributed environments. In this context, Multi-Agent Systems (MAS) have emerged as a key strategy for coordinating autonomous agents in complex tasks. The main objective of this article is to present the state of the art of MAS from the perspective of contemporary Artificial Intelligence, with particular emphasis on their integration with Large Language Models (LLMs) and the emerging role of the Model Context Protocol (MCP) as a facilitating element in this interaction. The methodology employed is qualitative, grounded in a narrative literature review guided by a systematic survey of publications in databases such as Google Scholar and Scopus. Additionally, Excel was used for the segmentation and critical analysis of the documents. The findings of this investigation indicate that MAS have become increasingly sophisticated in response to the growing complexity of distributed environments, particularly with the support of generative AI. The integration with LLMs and the emergence of MCP as a communication protocol play a central role in this progress. MCP enables interoperability and distributed autonomous decision-making, showing promise despite its early stage of adoption. Challenges remain, and future research should assess its effectiveness in more diverse scenarios. This study provides a conceptual foundation for the technical and ethical development of MAS in the era of AI.