In this study, we developed an MR-based upper limb assessment system that integrates a virtual demonstration hand with wearable sensors to capture kinematics during seven standardized tasks derived from the Fugl-Meyer Assessment (FMA) and Action Research Arm Test (ARAT). Ten healthy participants completed video-, VR-, and MR-based assessments, followed by a one-week retest of MR and VR conditions. Subjective experience was evaluated using multidimensional questionnaires, while the reliability of objective kinematic features was examined using intraclass correlation coefficients (ICC), standard error of measurement (SEM), and coefficient of variation (CV%).