Glycosylation changes are closely related to various diseases, including cancer. The quantitative analysis of site-specific glycans at proteomics scale remains challenging due to low glycopeptide spectra interpretation. Here, we present GlyPep-Quant, a tool for sensitive quantification and identification of site-specific glycans. Using a well-trained machine learning model, GlyPep-Quant quantified 25.1%-178.9% more site-specific glycans without missing values than pGlycoQuant, MSFragger-Glyco, and Skyline. To utilize identified information from previous large-scale dataset, an MS1 feature library-based "virtual match-between-runs" quantification scheme was developed, enabling over eightfold more site-specific glycan identification/quantification than conventional MS2-based methods. Enhanced coverage prompted the development of a glycoproteomic biomarker discovery method, involving calculation of site-specific glycan abundances ratios at the same glycosylation site, minimizing individual expression and experimental condition variability. Two pairs of site-specific glycan ratios on sites P01011-N127 and P08185-N96, were selected as high-performance biomarkers to classify gastric cancer (GC) from healthy controls (AUC > 0.95). Moreover, the two ratios performed well in distinguishing GC using an independent cohort by the library-based quantification strategy with diagnostic accuracy up to 85%. GlyPep-Quant is poised for broader glycoproteomic applications.