AGB stars are responsible for producing a variety of elements, including carbon, nitrogen, and the heavy elements produced in the slow neutron-capture process (s-elements). There are many uncertainties involved in modelling the evolution and nucleosynthesis of AGB stars, and this is especially the case at low metallicity, where most of the stars with high enough masses to enter the AGB have evolved to become white dwarfs and can no longer be observed. The stellar population in the Galactic halo is of low mass (≲0.85 M⊙) and only a few observed stars have evolved beyond the first giant branch. However, we have evidence that low-metallicity AGB stars in binary systems have interacted with their low-mass secondary companions in the past. The aim of this work is to investigate AGB nucleosynthesis at low metallicity by studying the surface abundances of chemically peculiar very metal-poor stars of the halo observed in binary systems. To this end we select a sample of 15 carbon- and s-element-enhanced metal-poor (CEMP-s) halo stars that are found in binary systems with measured orbital periods. With our model of binary evolution and AGB nucleosynthesis, we determine the binary configuration that best reproduces, at the same time, the observed orbital period and surface abundances of each star of the sample. The observed periods provide tight constraints on our model of wind mass transfer in binary stars, while the comparison with the observed abundances tests our model of AGB nucleosynthesis. For most of the stars in our sample, we find that an episode of efficient wind mass transfer, combined with strong angular momentum loss, has occurred in the past. In some cases we find discrepancies between the observed and modelled abundances even if we adopt a fine-tuned set of parameters in our binary evolution model. These discrepancies are probably caused by missing physical ingredients in our models of AGB nucleosynthesis and they provide indications of how to improve our knowledge of the process of nucleosynthesis in AGB stars.