The management of metastatic breast cancer (mBC) relies on tissue-based immunohistochemical subtypes. However, biopsies are invasive, may not capture metastatic heterogeneity, and subtypes can change over time under treatment pressure. Here, we developed cell-free DNA (cfDNA) methylation signatures for minimally invasive BC detection, distinction, and estrogen receptor (ER) status classification. Peripheral blood plasma methylomes were analyzed from 79 patients with mBC spanning ER+/HER2- (n=45), HER2+ (n=13), and triple-negative BC (TNBC; n=21). To derive tissue-informed BC and ER-specific features, public 450K methylation array data (n=9730) were leveraged, and features were selected using generalized linear models via elastic net regularization (GLMnet) with cross-validation. The tissue-informed features were translated to cell-free methylated DNA immunoprecipitation and sequencing (cfMeDIP-seq), and the final signatures were validated across a compendium of cfMeDIP-seq profiles (n=713) spanning over ten cancer types. Across training, validation, and external test cohorts, the signatures demonstrated high accuracy for BC detection versus controls, distinction from multiple other malignancies, and ER status classification. Performance generalized across independent cfMeDIP-seq cohorts and reflected tumor fraction. The sensitivity was reduced in samples with low tumor fractions and bone-only disease, while remaining informative for typical tumor fractions observed in the metastatic setting. Promoter-proximal signature regions provided biological insight into tumor phenotypes. This tissue-anchored, platform-translatable framework demonstrates the feasibility of accurate, reproducible cfDNA methylation-based molecular classification in mBC.