High-resolution MRF including a multi-pool saturation transfer model requires a huge dictionary or training dataset simulated with Bloch equations. Furthermore, intensive Bloch simulation tasks are inevitable for MRF schedule optimization. In this study, we developed a deep-learning-based ultrafast Bloch simulator and a recurrent neural network (RNN) for semi-solid macromolecular magnetization transfer contrast (MTC) MRF reconstruction. For the MRF training dataset generation, the deep-learning Bloch simulator required ~200x less time than a conventional Bloch simulation. A test-retest study showed excellent reliability of the tissue parameter quantification using the proposed RNN framework.