Understanding how giant planets form requires observational input from directly imaged protoplanets. We used VLT/NACO and VLT/SPHERE to search for companions in the transition disc of 2MASS J19005804-3645048 (hereafter CrA-9), an accreting M0.75 dwarf with an estimated age of 1–2 Myr. We found a faint point source at ∼0.7” separation from CrA-9 (∼108 au projected separation). Our 3-epoch astrometry rejects a fixed background star with a 5σ significance. The near-IR absolute magnitudes of the object point towards a planetary-mass companion. However, our analysis of the 1.0–3.8μm spectrum extracted for the companion suggests it is a young M5.5 dwarf, based on both the 1.13-μm Na index and comparison with templates of the Montreal Spectral Library. The observed spectrum is best reproduced with high effective temperature ($3057^{+119}_{-36}$K) BT-DUSTY and BT-SETTL models, but the corresponding photometric radius required to match the measured flux is only $0.60^{+0.01}_{-0.04}$ Jovian radius. We discuss possible explanations to reconcile our measurements, including an M-dwarf companion obscured by an edge-on circum-secondary disc or the shock-heated part of the photosphere of an accreting protoplanet. Follow-up observations covering a larger wavelength range and/or at finer spectral resolution are required to discriminate these two scenarios.