Due to the highly heterogeneous nature of hepatocellular carcinoma (HCC), the accurate diagnosis of HCC during the early phase of development is still a challenging task. Therefore, the further development of novel diagnostic methods by discovering new biomarkers is required to improve the rate of HCC diagnosis in the early phase. In this work, an oxygen-modified three-dimensional interconnected porous carbon probe is designed and fabricated to profile the differences of N-glycans in human serum from health controls (H) and patients with hepatic dysfunction (HD) and HCC for the discovery of new biomarkers with HCC development. Excitingly, we discovered that the expression levels of 12 serum N-glycans were gradually increased from H to patients with HD and eventually to patients with HCC. Moreover, two machine learning models established based on these 12 serum N-glycans reached a satisfactory accuracy for predicting HCC development where the receiver operating characteristic curve arrived above 0.95 for distinguishing healthy controls and patients with liver diseases (HD or HCC) and the ROC curve arrived at 0.85 for distinguishing HD and HCC. Our work not only developed a new method for the large-scale characterization of serum N-glycans but also provided valuable guidance for accurate and highly sensitive diagnosis of early liver cancer development in a non-invasive manner.