On average, two tropical storms or cyclones enter the Mozambique Channel between the African mainland and Madagascar each year. Their impact can be devastating. The tropical cyclone Idai, which hit land in Mozambique in 2019, was one of the deadliest storms on record in the Southern Hemisphere. Previous studies have found that the tracks and strengths of tropical storms and cyclones are difficult to predict more than a few days ahead. An extension of this forecast horizon would be crucial for enabling authorities to take precautionary actions. Here, the ability of a state-of-the-art ensemble prediction model to predict Idai and 38 other tropical systems is assessed. It is found that the minimum sea level pressure (SLP) associated with Idai was only skilfully predicted at lead times up to three days. When considering all the systems, less than a quarter of the ensemble members predicted lower minimum SLP than the MERRA-2 reanalysis at lead times of five days and longer. However, several variables are found to be useful precursors of tropical storms and cyclones, and some of these are skilfully predicted at long lead times. In particular, area-averaged anomalies of geopotential height and specific humidity at 500 hPa and SLP one week before the passage of storms are significantly correlated with minimum SLP anomalies during the storms. As these precursor variables are skilfully predicted by the model at lead times of 10–12 days, it should be possible to include their forecasted values in hybrid statistical–dynamical prediction systems at lead times beyond a few days. An additional interesting finding is that warm sea surface temperature anomalies and weak vertical wind shear, which are generally considered to be favourable for tropical storms and cyclones, do not qualify as precursors of the systems investigated here.