COMSOL-MCP: An open-source model context protocol interface for AI-assisted multiphysics simulation

Multiphysics simulation software remains difficult to use through natural language because simulation workflows combine domain knowledge, stateful software sessions, geometry construction, entity selection, meshing, solving, and post-processing. We present COMSOL-MCP, an open-source implementation of the Model Context Protocol (MCP) that exposes COMSOL Multiphysics operations to large-language-model agents through structured tool calls. The system provides 76 tools covering session management, model construction, parameters, geometry, physics setup, meshing, studies, results, and knowledge access. A knowledge component combines deterministic embedded physics guides with semantic retrieval from 36,798 COMSOL documentation chunks spanning 11 physics domains. Rather than claiming validated autonomous simulation, this paper reports a software architecture and an exploratory micromixer CFD trace study. The trace shows that an agent can execute many operational steps without COMSOL GUI interaction, including model creation, geometry construction through a fallback script, mesh generation, solver execution, and result extraction. It also exposes a consequential failure: boundary-condition entity assignment was incorrect, producing a solver-converged but physically invalid flow field. We therefore position COMSOL-MCP as a reusable interface and diagnostic prototype, not as a fully autonomous simulation system. The case study identifies spatial grounding, topology-aware entity selection, and rigorous multi-problem benchmarking as necessary next steps for reliable AI-assisted multiphysics workflows.