Multiple myeloma (MM) is a hematological malignancy characterised by the clonal proliferation of abnormal plasma cells within the bone marrow (BM). Despite advances in treatment that have improved survival, the disease remains incurable. MM diagnosis requires invasive bone marrow biopsy to quantify the percentage of malignant plasma cells. In this study, the potential of extracellular vesicles (EVs) as a non-invasive liquid biopsy for MM diagnosis and staging was investigated. The proteomic content of patient-derived EVs was profiled via mass spectrometry from peripheral blood and bone marrow of 33 MM patients and 12 healthy donors. Biomarker signatures were identified using supervised machine learning to predict asymptomatic MM, progression to symptomatic MM, and relapse. The analysis identified a six-protein biomarker signature, comprising APOC1, KRT78, ALAD, S100A7, LGALS1, and CD226 forming four optimal logistic regression diagnostic MM models with predictive accuracies of >85%. The performance of the identified proteins, supports their potential as a minimally invasive EV-based liquid biopsy in MM diagnosis and monitoring.