We examine the cosmological constraints that can be achieved with a galaxy cluster survey with the future CORE space mission. Using realistic simulations of the millimeter sky, produced with the latest version of the Planck Sky Model, we characterize the CORE cluster catalogues as a function of the main mission performance parameters. We pay particular attention to telescope size, key to improved angular resolution, and discuss the comparison and the complementarity of CORE with ambitious future ground-based CMB experiments that could be deployed in the next decade. A possible CORE mission concept with a 150 cm diameter primary mirror can detect of the order of 50,000 clusters through the thermal Sunyaev-Zeldovich effect (SZE). The total yield increases (decreases) by 25% when increasing (decreasing) the mirror diameter by 30 cm. The 150 cm telescope configuration will detect the most massive clusters (>1014 M⊙) at redshift z>1.5 over the whole sky, although the exact number above this redshift is tied to the uncertain evolution of the cluster SZE flux-mass relation; assuming self-similar evolution, CORE will detect 0~ 50 clusters at redshift z>1.5. This changes to 800 (200) when increasing (decreasing) the mirror size by 30 cm. CORE will be able to measure individual cluster halo masses through lensing of the cosmic microwave background anisotropies with a 1-σ sensitivity of 4×1014 M⊙, for a 120 cm aperture telescope, and 1014 M⊙ for a 180 cm one. From the ground, we estimate that, for example, a survey with about 150,000 detectors at the focus of 350 cm telescopes observing 65% of the sky would be shallower than CORE and detect about 11,000 clusters, while a survey with the same number of detectors observing 25% of sky with a 10 m telescope is expected to be deeper and to detect about 70,000 clusters. When combined with the latter, CORE would reach a limiting mass of M500 ~ 2−3 × 1013 M⊙ and detect 220,000 clusters (5 sigma detection limit). Cosmological constraints from CORE cluster counts alone are competitive with other scheduled large scale structure surveys in the 2020's for measuring the dark energy equation-of-state parameters w0 and wa (σw0=0.28, σwa=0.31). In combination with primary CMB constraints, CORE cluster counts can further reduce these error bars on w0 and wa to 0.05 and 0.13 respectively, and constrain the sum of the neutrino masses, Σ mν, to 39 meV (1 sigma). The wide frequency coverage of CORE, 60-600 GHz, will enable measurement of the relativistic thermal SZE by stacking clusters. Contamination by dust emission from the clusters, however, makes constraining the temperature of the intracluster medium difficult. The kinetic SZE pairwise momentum will be extracted with 0S/N=7 in the foreground-cleaned CMB map. Measurements of TCMB(z) using CORE clusters will establish competitive constraints on the evolution of the CMB temperature: (1+z)1−β, with an uncertainty of σβ lesssim 2.7× 10−3 at low redshift (z lesssim 1). The wide frequency coverage also enables clean extraction of a map of the diffuse SZE signal over the sky, substantially reducing contamination by foregrounds compared to the Planck SZE map extraction. Our analysis of the one-dimensional distribution of Compton-y values in the simulated map finds an order of magnitude improvement in constraints on σ8 over the Planck result, demonstrating the potential of this cosmological probe with CORE.