Stroke often results in hemiparesis, impairing the patient's motor abilities and leading to upper extremity motor deficits that require long-term training and assessment. However, existing methods for assessing patients' motor function rely on clinical scales that require experienced physicians to guide patients through target tasks during the assessment process. This process is not only time-consuming and labor-intensive, but the complex assessment process is also uncomfortable for patients and has significant limitations. For this reason, we propose a serious game that automatically assesses the degree of upper limb motor impairment in stroke patients. Specifically, we divide this serious game into a preparation stage and a competition stage. In each stage, we construct motor features based on clinical a priori knowledge to reflect the ability indicators of the patient's upper limbs. These features all correlated significantly with the Fugl-Meyer Assessment for Upper Extremity (FMA-UE), which assesses motor impairment in stroke patients. In addition, we design membership functions and fuzzy rules for motor features in combination with the opinions of rehabilitation therapists to construct a hierarchical fuzzy inference system to assess the motor function of upper limbs in stroke patients. In this study, we recruited a total of 24 patients with varying degrees of stroke and 8 healthy controls to participate in the Serious Game System test. The results show that our Serious Game System was able to effectively differentiate between controls, severe, moderate, and mild hemiparesis with an average accuracy of 93.5%.