Background Maternal mortality in Africa has changed little since 1990. We developed a mathematical model with the aim to assess whether improved community-based access to life-saving drugs, to augment a core programme of health-facility strengthening, could reduce maternal mortality due to post-partum haemorrhage or sepsis. Methods We developed a mathematical model by considering the key events leading to maternal death from post-partum haemorrhage or sepsis after delivery. With parameter estimates from published work of occurrence of post-partum haemorrhage and sepsis, case fatality and the effectiveness of drugs, we used this model to estimate the effect of three potential packages of interventions: 1) health-facility strengthening; 2) health-facility strengthening combined with improved drug provision via antenatal-care appointments and community health workers; and 3) all interventions in package two combined with improved community-based drug provision via female volunteers in villages. The model was applied to Malawi and sub-Saharan Africa. Findings In the implementation of the model, the lowest risk deliveries were those in health facilities. With the model we estimated that of 2860 maternal deaths from post-partum haemorrhage or sepsis per year in Malawi, intervention package one could prevent 210 (7%) deaths, package two 720 (25%) deaths, and package three 1020 (36%) deaths. In sub-Saharan Africa, we estimated that of 182 000 of such maternal deaths per year, these three packages could prevent 21300 (12%), 43 800 (24%), and 59 000 (32%) deaths, respectively. The estimated effect of community-based drug provision was greatest for the poorest women. Interpretation Community provision of misoprostol and antibiotics to reduce maternal deaths from post-partum haemorrhage and sepsis could be a highly effective addition to health-facility strengthening in Africa. Investigation of such interventions is urgently needed to establish the risks, benefits, and challenges of widespread implementation. Funding Institute of Child Health and Faculty of Mathematical and Physical Sciences, University College London, and a donation from John and Ann-Margaret Walton