No. You have removed information in the process of estimating the density. The density
function does not store its input the way, for example, ecdf
does.
Edit. I can perhaps soften that a bit. You could use the information in density to reconstruct a rough estimate of the original values. You would need to know the sample size, which which is recorded in the object returned, and then use the estimate to generate new data. Integrate the density, split the cumulative probability function equally along its range of support and you could construct an empiric quantile estimate.