Cataloging astrophysical sources is a fundamental operation in astronomy. While simple in principle, cataloging becomes more complicated for images with low signal to noise and degeneracies across different emission components. We analyze Chandra Deep Field - South (CDF-S) data using a novel method called probabilistic cataloging, which extracts information by sampling from the catalog space, i.e. the space of different point source configurations consistent with a given image. By employing a reversible-jump Markov Chain Monte Carlo (RJMCMC) sampling method, we are able to infer the flux and color distribution for Active Galactic Nuclei (AGN) in the region. We are also able to infer the number of AGN by marginalizing over faint members below the detection threshold. To validate our method, we use simulated deep Chandra exposures and show that the isotropic background emission can be constrained at the 10% level. This result takes into account its covariance with unresolved AGN and the particle background of Chandra. We then present results of probabilistic cataloging applied to the CDF-S 2Ms, 4Ms, and 7Ms datasets in order to evaluate the fidelity of our method. Furthermore, by incorporating auxiliary redshift information from the COMBO Survey within our framework, we present the first three-dimensional probabilistic catalog of AGN and discuss possible implications for AGN synthesis models.