Mapping simulation fields between disparate discretizations is a critical component of multiscale and multiphysics workflows, including fluid–structure interactions, adaptive remeshing, and neutral particle transport in fusion reactor modeling. Many commonly used field transfer techniques, such as interpolation and radial basis functions, rely solely on pointwise field evaluations but generally fail to conserve integral quantities such as mass or energy. In contrast, mesh intersection–based methods provide high-fidelity conservative transfer but require explicit access to full source and target discretizations, limiting their applicability in black-box workflows. This talk presents a GPU-accelerated Monte Carlo field transfer methodology that achieves conservative transfer without requiring mesh intersections or explicit mesh connectivity, relying only on pointwise field queries. Accuracy, conservation, and performance on GPUs are evaluated and compared against radial basis functions and traditional mesh-intersection-based transfer techniques. Results demonstrate that the proposed method achieves conservation and accuracy that is comparable to mesh-intersection methods while providing a scalable and flexible alternative for conservative coupling in large-scale multiscale and multiphysics simulations. This method has been implemented in the Parallel Coupler for Multimodel Simulations (PCMS), a multiphysics coupling library designed to support the wide range of physics-based coordinate systems and high-dimensional discretizations encountered in fusion reactor modeling. The talk will present representative fusion coupling examples enabled by PCMS and discuss extensions of the methodology toward five-dimensional conservative field transfer.