Dynamic analysis is fundamental to enhancing mechanical system performance, and the efficiency of multibody system dynamics (MSD) modeling is largely determined by the underlying methodology. This paper proposes a novel direct differentiation sensitivity analysis method based on the Reduced Multibody System Transfer Matrix Method (RMSTMM). By integrating direct sensitivity analysis with a recursive element-level formulation, the proposed method avoids the derivation of global dynamic equations, ensuring high computational efficiency and programmability. Furthermore, the approach is extended to handle closed-loop multibody systems by introducing constraint equations at cut joints and establishing corresponding recursive sensitivity formulations. Numerical examples validate the method’s accuracy and effectiveness. Compared with traditional Lagrangian-based methods such as the Direct Differentiation Method (DDM) and the Adjoint Variable Method (AVM), the proposed method achieves lower matrix order, faster computation, and improved numerical stability, making it well-suited for large-scale or topologically complex systems.