Efficient and sustainable models of Corona Virus Disease of 2019 (COVID‑19) screening among healthcare workers (HCWs) such as self-testing are essential for averting transmission within and outside health care facilities. We compared the number of confirmed early COVID‑19 positive diagnoses (defined as COVID‑19 cases diagnosed within five days of symptoms onset) among HCW in self-testing arm using Antigen Rapid Diagnostic Test kits (Ag-RDT) and standard of care arm (SOC) who were offered COVID‑19 testing if they presented at the clinic with any COVID‑19 symptoms. Twelve primary healthcare facilities in Blantyre, Malawi, were purposively allocated (1:1) aiming for geographical and size balance in a 2-arm non-randomised cluster trial (ISRCTN: 17596113), available from https://doi.org/10.1186/ISRCTN17596113. Arm-1 was SOC and Arm-2 was COVID 19 self-testing (C19ST). HCWs in the C19ST arm had twice-weekly COVID‑19 Ag-RDT self-testing. The primary outcome compared by arm the harmonic mean number of early COVID‑19 positive diagnoses among HCWs. Analysis was by intention-to-treat using cluster-level summaries and t-test, with adjustment for imbalance. Participation was 99.8% among eligible HCWs across all facilities (1081/1083). Of the 1081 participating, 612 (56.6%) and 469 (43.4%) were in SOC and C19ST arm, respectively. Mean age was 35.5y (sd: 9.3); 183/612 (29.9%) in SOC were male, compared to 166/469 (35.3%) in C19ST; overall prior vaccination was 80.0% with no difference between SOC, (81.5%) and C19ST (78.0%). Follow-up at exit (12 weeks) was high (SOC: [94%]; C19ST: [87.6%]) and a harmonic mean of 1 and 4 HCWs had early COVID‑19 diagnosis in SOC and C19ST arms, respectively. COVID‑19 self-testing using Ag-RDTs provided a safe, quick, and reliable model for identifying early-onset symptomatic and asymptomatic COVID‑19 positive HCW. Self-testing was feasible to integrate for routine screening among HCW potentially reducing disruption to health services. This model has potential for wide scale up programmatically especially in resource-constrained settings.