Context. Gaia Data Release 2 (DR2) provides a unique all-sky catalogue of 550 737 variable stars, of which 151 761 are long-period variable (LPV) candidates with G variability amplitudes larger than 0.2 mag (5–95% quantile range). About one-fifth of the LPV candidates are Mira candidates, the majority of the rest are semi-regular variable candidates. For each source, G, G_BP, and G_RP photometric time-series are published, together with some LPV-specific attributes for the subset of 89 617 candidates with periods in G longer than 60 days. Aims. We describe this first Gaia catalogue of LPV candidates, give an overview of its content, and present various validation checks. Methods. Various samples of LPVs were used to validate the catalogue: a sample of well-studied very bright LPVs with light curves from the American Association of Variable Star Observers that are partly contemporaneous with Gaia light curves, a sample of Gaia LPV candidates with good parallaxes, the All-Sky Automated Survey for Supernovae catalogue of LPVs, and the Optical Gravitational Lensing Experiment (OGLE) catalogues of LPVs towards the Magellanic Clouds and the Galactic bulge. Results. The analyses of these samples show a good agreement between Gaia DR2 and literature periods. The same is globally true for bolometric corrections of M-type stars. The main contaminant of our DR2 catalogue comes from young stellar objects (YSOs) in the solar vicinity (within ~1 kpc), although their number in the whole catalogue is only at the percent level. A cautionary note is provided about parallax-dependent LPV attributes published in the catalogue. Conclusions. This first Gaia catalogue of LPVs approximately doubles the number of known LPVs with amplitudes larger than 0.2 mag, despite the conservative candidate selection criteria that prioritise low contamination over high completeness, and despite the limited DR2 time coverage compared to the long periods characteristic of LPVs. It also contains a small set of YSO candidates, which offers the serendipitous opportunity to study these objects at an early stage of the Gaia data releases.