Accurate and precise measurements of the Hubble constant are critical for testing our current standard cosmological model and revealing possibly new physics. With Hubble Space Telescope (HST) imaging, each strong gravitational lens system with measured time delays can allow one to determine the Hubble constant with an uncertainty of ∼7 per cent. Since HST will not last forever, we explore adaptive-optics (AO) imaging as an alternative that can provide higher angular resolution than HST imaging but has a less stable point spread function (PSF) due to atmospheric distortion. To make AO imaging useful for time-delay-lens cosmography, we develop a method to extract the unknown PSF directly from the imaging of strongly lensed quasars. In a blind test with two mock data sets created with different PSFs, we are able to recover the important cosmological parameters (time-delay distance, external shear, lens-mass profile slope, and total Einstein radius). Our analysis of the Keck AO image of the strong lens system RXJ 1131-1231 shows that the important parameters for cosmography agree with those based on HST imaging and modelling within 1σ uncertainties. Most importantly, the constraint on the model time-delay distance by using AO imaging with 0.09 arcsec resolution is tighter by ∼50 per cent than the constraint of time-delay distance by using HST imaging with 0.09 arcsec when a power-law mass distribution for the lens system is adopted. Our PSF reconstruction technique is generic and applicable to data sets that have multiple nearby point sources, enabling scientific studies that require high-precision models of the PSF.