Refined constraints on chameleon theories are calculated for atom-interferometry experiments, using a numerical approach consisting in solving for a four-region model the static and spherically symmetric Klein-Gordon equation for the chameleon field. By modeling not only the test mass and the vacuum chamber but also its walls and the exterior environment, the method allows to probe new effects on the scalar field profile and the induced acceleration of atoms. In the case of a weakly perturbing test mass, the effect of the wall is to enhance the field profile and to lower the acceleration inside the chamber by up to one order of magnitude. In the thin-shell regime, results are found to be in good agreement with the analytical estimations, when measurements are realized in the immediate vicinity of the test mass. Close to the vacuum chamber wall, the acceleration becomes negative and potentially measurable. This prediction could be used to discriminate between fifth-force effects and systematic experimental uncertainties, by doing the experiment at several key positions inside the vacuum chamber. For the chameleon potential $V(\phi) = \Lambda^{4+\alpha} / \phi^\alpha$ and a coupling function $A(\phi) = \exp(\phi /M)$, one finds $M \gtrsim 7 \times 10^{16} \mathrm{GeV}$, independently of the power-law index. For $V(\phi) = \Lambda^4 (1+ \Lambda/ \phi)$, one finds $M \gtrsim 10^{14} \mathrm{GeV}$. A sensitivity of $a\sim 10^{-11} \mathrm{m/s^2} $ would probe the model up to the Planck scale. Finally, a proposal for a second experimental set-up, in a vacuum room, is presented. In this case, Planckian values of $M$ could be probed provided that $a \sim 10^{-10} \mathrm{m/s^2}$, a limit reachable by future experiments. Our method can easily be extended to constrain other models with a screening mechanism, such as symmetron, dilaton and f(R) theories.