Metabolic glycan labeling (MGL) followed by bioorthogonal chemistry provides a powerful tool for tumor imaging and therapy. However, selectively metabolic labeling of cells or tissues of interest remains a challenge. Particularly, owing to tumor heterogeneity including tumor subtypes and interpatient heterogeneity, it is far more difficult to realize tumor-cell-selective metabolic labeling for precise diagnosis. Inspired by nature, we designed azidosugar-functionalized metal-organic frameworks camouflaged with cancer cell membranes to accomplish cancer-cell-selective MGL in vivo. With abundant receptors, this biomimetic platform not only selectively targets homotypic cells but also realizes different breast cancer subtype-selective MGL. Moreover, the endo/lysosomal-escaped ZIF-8 can make azidosugar escape from lysosomes and accelerate its metabolic incorporation. This strategy also takes advantage of cancer-tissue-derived cell membranes, which may have huge potential for personalized diagnosis and therapy.