The article introduces DiScoFormer (Density and Score Transformer), a model that estimates both the density and the score of a distribution from a set of data points in a single forward pass without retraining. Many problems in machine learning and the sciences involve recovering a distribution from data points, which requires estimating density (a smooth histogram) and score (the gradient of the log-density, which points toward more probable regions). Score drives diffusion-based generative models like Stable Diffusion and DALL-E, as well as Bayesian sampling and particle simulations. Current tools face a trade-off: kernel density estimation (KDE) needs no training and applies to any distribution but its accuracy falls off sharply as dimensionality grows, while neural score-matching models stay accurate in high dimensions but must be retrained for each new distribution. DiScoFormer uses stacked transformer blocks with cross-attention to evaluate density and score at any point. It has a shared backbone with two output heads for density and score, leveraging their mathematical relationship: score is the gradient of the logarithm of density. This coupling creates a label-free consistency loss, allowing DiScoFormer to adapt to out-of-distribution inputs at inference by taking a few gradient steps on that loss, with no ground-truth density or score required.