We present MuscleMimic, an open-source framework for scalable motion imitation learning with physiologically realistic, muscle-actuated humanoids. MuscleMimic introduces two musculoskeletal models — BimanualMuscle (76 joints, 126 muscles) for upper-body manipulation and MyoFullBody (123 joints, 416 muscles) for full-body locomotion and manipulation — and leverages JAX-based GPU-accelerated simulation to train on thousands of human motions at scale. Trained across 8,192 parallel environments for 4.9 billion timesteps, our policies achieve a mean correlation of 0.92 and 0.94 for walking kinematics, with muscle activation patterns validated against EMG recordings.