Stroke patients commonly suffer from motor impairments, leading to a heightened interest in the application of rehabilitation robots and serious games for motor re-learning therapies. A wealth of research has established that the integration of motor and cognitive training modalities can substantially enhance the rehabilitation process. Nevertheless, accurately assessing patients’ motor and cognitive statuses remains a significant challenge for current methodologies, thereby impeding real-time monitoring and adaptive adjustments to the rehabilitation regimen. In this work, we introduce a personalized motor-cognitive training framework anchored in serious gaming. Leveraging dynamic system theory, our framework is capable of extracting interpretable dynamic characteristics from the human-robot interaction, enabling a precise evaluation of subjects’ motor skill abilities. The training parameters can be dynamically tuned with the assessed skill levels, which supports multidimensional personalized rehabilitation. Building upon this approach, we have engineered an innovative motor-cognitive rehabilitation platform that synergizes serious games with robotic movement interaction. We have conducted experiments to validate the platform’s efficacy, demonstrating its potential to advance the field of post-stroke rehabilitation.