Multi-agent AI and MCP servers automate power grid simulations

Power grid operators face increasingly complex simulations to manage transmission stability, renewable integration, and contingency planning. In a new position paper accepted at the IJCAI AISE 2026 workshop, researchers Jérôme Picault and Clément Goubet from RTE (France's transmission system operator) propose using multi-agent AI systems and the Model Context Protocol (MCP) to orchestrate these studies. The paper outlines how large language models can act as orchestrators, delegating simulation tasks to specialized agents that interact with numerical solvers through standardized tool calls. This approach promises to make grid studies more interactive, auditable, and scalable, while keeping human experts in the loop for critical decisions.

The key technical contribution is pypowsybl-mcp, an MCP-based interface that exposes selected capabilities of the pypowsybl power-system simulation library to AI agents. This allows agents to set up simulations, run load-flow analyses, retrieve detailed results, and chain multiple study steps autonomously. The researchers also define principles for multi-agent workflows with human supervision, including mechanisms for verification, rollback, and escalation. An evaluation strategy combining technical metrics (simulation accuracy, task completion time) with practitioner feedback is proposed. By standardizing tool integration through MCP, the work aims to bridge the gap between AI reasoning and industrial-grade power engineering tools.