With the growing needs on rehabilitation of the mild cognitive impairment (MCI) users group and the advantages of virtual reality (VR) technologies in cognitive training, the development of VR-based rehabilitation training methods has become a hot spot recently. However, the challenges in accurately measuring users’ needs and quantifying training system efficacy are still not well resolved, especially for longitudinal tracking. In this study, a VR-based cognitive training and evaluation system is designed and implemented, targeting at fulfilling the rehabilitation needs of MCI users. It evaluates the impact of longitudinal VR-based training on MCI users with a number of feedback methodologies including brain activation indicators, brain network connectivity indicators, behavioral indicators and the Montreal Cognitive Assessment (MoCA) scale scores, extracted from multi-modal data collected while training. A two-month longitudinal tracking ergonomics experiment was conducted to validate the usability of the feedback methodologies and to explore the influence of the training duration on the rehabilitation efficacy. The results showed that our proposed VR-based cognitive training and evaluation system had a positively significant impact on the rehabilitation of the MCI group. Meanwhile, the multi-source feedbacks can also help the updates and iterations of VR-based rehabilitation training systems. Finally, this study provides guidance for the selection of rehabilitation cycles and emphasizes the importance of quantitative studies with longitudinal follow-up in assessing rehabilitation efficacy.