The design specification of future high-field accelerator magnets require innovation in and integration of multiple disciplines. Numerical models underpin each step of the design. In this paper, we present a methodology for collaborative modeling based on relevant concepts of model-based systems engineering. The methodology is composed of three pillars: encapsulated computing environments with service query interface, model notebooks with auto-generated model views, and model query interface with results caching. The methodology aims at a de-centralized approach to multi-model and multi-scale collaboration that maintains the cornerstones of traceability and reproducibility. It is demonstrated with a multi-model coil design optimization of a high-field superconducting magnet.