This paper describes a way to connect modern AI agents to industry power‑system simulators so the agents can run routine study tasks. The researchers built a custom Model Context Protocol (MCP) server that exposes Siemens PTI PSS®E functions. The goal is for AI agents to orchestrate validated engineering tools instead of replacing them.

To test the idea, the team implemented two agentic pathways. One used the OpenAI Agents software development kit (SDK) and the other used Claude Code via a command‑line interface. Both implementations used reusable “skills” (procedures and templates), subagents, MCP tools, links to data repositories, and local shell or Python steps. The MCP server in this work exposes 21 PSS®E functions for tasks like case management, power‑flow solution, dynamic simulation, and result extraction.

At a high level the system works like this: an orchestration agent receives a study objective and breaks it into subtasks. It delegates subtasks to task agents that follow skill files encoding required inputs, sequences of tool calls, validation checks, and failure rules. One specific workflow tested was a model‑validation “play‑in” approach, where recorded measurements (for example voltages and frequencies) drive a simulation and the simulated power responses are compared to the measurements. The team judged success by whether tasks completed, whether outputs were accurate, and how often human experts had to step in.

Both frontier‑model implementations successfully executed representative study tasks on public data sets. The authors report that agentic systems can greatly accelerate the dynamic simulation parts of transmission planning studies when they call industry‑grade simulators. If adopted, such systems could let engineers spend less time on tool operation and more on designing scenarios and interpreting results.