Background: Hepatocellular carcinoma (HCC), a prevalent malignancy and a leading cause of cancer-related mortality, often exhibits a grim prognosis due to recurrence post-surgical treatment. Minimal residual disease (MRD) is deemed a crucial factor contributing to early relapse following radical surgery. Previous studies have indicated that circulating tumor DNA (ctDNA) may serve as a promising biomarker for evaluating MRD in various cancers. However, its applicability in liver cancer remains unclear. Methods: We prospective included 136 patients during 2019 to 2021 who underwent curative-intention surgery for HCC in Zhongshan hospital. Plasma samples (n=625) were collected prior to surgery, 1-month post-surgery, and during subsequent surveillance. Patient-specific somatic mutations were identified through whole-exome sequencing of tumor tissue. For each patient, 16 patient-specific somatic mutations were selected and employed to assess ctDNA through ultra-deep sequencing of plasma DNA. Results: The median follow-up period was 24 months. At the last follow-up, 45 (33.1%) patients relapsed. The preoperative ctDNA concentration was elevated in patients who experienced tumor relapse after surgical treatment compared to those didn’t relapse (median: 125 vs.8.4 mean tumor molecules/mL, P<0.001). Most of patients (80.9%) demonstrated a significant decrease in ctDNA concentration to negative levels after curative surgery. Those maintaining a positive postoperative ctDNA status faced a higher recurrence risk (HR=11.5, P<0.001). Throughout longitudinal follow-up, positive longitudinal ctDNA (positive in any point) was significantly associated with a higher recurrence risk (HR=69.3, P<0.001). The area under the curve of longitudinal ctDNA for predicting recurrence was 0.956, boasting a sensitivity of 93.3% and specificity of 97.8%. The median interval from the first positive longitudinal ctDNA to clinical recurrence confirmed by imaging test was 4 months. The predictive performance of longitudinal ctDNA for recurrence surpassed that of alpha-fetoprotein (0.956 vs. 0.680, P<0.001). Patients with positive longitudinal ctDNA experienced benefits from adjuvant therapy (HR=0.4, P=0.019). Moreover, a decrease or stability of ctDNA after adjuvant therapy indicated treatment response and delayed clinical relapse (HR=0.3, P=0.00012). Conclusions: Tumor-informed ctDNA proves to be a reliable indicator for MRD detection after surgical treatment for HCC, enabling the prediction and anticipation of recurrence before clinical manifestation. It plays a crucial role in guiding the application of adjuvant therapy and assessing treatment efficacy, thereby facilitating postoperative management and enhancing the precision of medical interventions.