ABSTRACT The finding of active galactic nuclei (AGNs) in dwarf galaxies has important implications for galaxy evolution and supermassive black hole formation models. Yet, how AGNs in dwarf galaxies form is still debated, in part due to scant demographics. We use the MaNGA survey, comprising ∼10 000 galaxies at z < 0.15, to identify AGN dwarf galaxies using a spaxel-by-spaxel classification in three spatially-resolved emission-line diagnostic diagrams (the [N ii]-, [S ii]-, and [O i]-BPT) and the WHAN diagram. This yields a sample of 664 AGN dwarf galaxies, the largest to date, and an AGN fraction of $\sim 20~{{\ \rm per\ cent}}$ that is significantly larger than that of single-fibre-spectroscopy studies (i.e. $\sim 1~{{\ \rm per\ cent}}$). This can be explained by the lower bolometric luminosity (<1042 erg s−1) and accretion rate (sub-Eddington) of the MaNGA AGN dwarf galaxies. We additionally identify 1176 SF-AGNs (classified as star-forming in the [N ii]-BPT but as AGNs in the [S ii]- and [O i]-BPT), 122 Composite, and 173 LINER sources. The offset between the optical centre of the galaxy and the median position of the AGN spaxels is more than 3 arcsec for ∼62% of the AGNs, suggesting that some could be off-nuclear. We also identify seven new broad-line AGNs with log MBH = 5.0–5.9 M⊙. Our results show how integral-field spectroscopy is a powerful tool for uncovering faint and low-accretion AGNs and better constraining the demographics of AGNs in dwarf galaxies.