Hepatocellular carcinoma (HCC) is a highly lethal malignancy with frequent recurrence after curative-intent surgery. Current detection and monitoring methods rely heavily on imaging, and genomic strategies are limited by access to tumor tissue and the challenges of mutational heterogeneity. Therefore, a cost-effective noninvasive approach with sufficient sensitivity, specificity, and high accuracy is needed to enhance HCC management. In this study, we evaluated the clinical potential of cfMeDIP-seq for HCC detection and postoperative monitoring. A total of 236 cfDNA samples were collected at surgery (b-HCC, n=89) and during follow-up (f-HCC, n=112) from 89 HCC patients undergoing liver transplantation (n=57) or resection (n=32), along with 35 healthy controls (CTL). cfMeDIP-seq was performed, followed by machine learning to develop an HCC-specific classifier in a discovery cohort (52 b-HCC vs. 35 CTL), which was subsequently tested in a validation cohort of 37 patients. An HCC methylation score (HMS) was assigned to reflect the probability of a sample containing HCC-derived cfDNA, and relationships between HMS and clinical variables were assessed. The classifier identified HCC with 97% sensitivity and 99% specificity in the discovery cohort and 97% accuracy in the validation cohort. Baseline HMS >0.9 was associated with higher recurrence risk (HR 3.43, 95% CI 1.30-9.06, p=0.013), and HMS decreased by 3-44% (median 17%) within 13 weeks post-surgery. HMS trajectories diverged for recurrent and non-recurrent patients, with an increase in HMS indicating clinical recurrence, and the HMS was independent of other clinicopathologic variables. These findings demonstrate that tumor-agnostic cfDNA methylomes can accurately detect HCC and predict recurrence after liver resection or transplantation, suggesting that this approach may have important implications for HCC diagnosis, treatment, and monitoring.