Primordial features, in particular oscillatory signals, imprinted in the primordial power spectrum of density perturbations represent a clear window of opportunity for detecting new physics at high-energy scales. Future spectroscopic and photometric measurements from the $Euclid$ space mission will provide unique constraints on the primordial power spectrum, thanks to the redshift coverage and high-accuracy measurement of nonlinear scales, thus allowing us to investigate deviations from the standard power-law primordial power spectrum. We consider two models with primordial undamped oscillations superimposed on the matter power spectrum, one linearly spaced in $k$-space the other logarithmically spaced in $k$-space. We forecast uncertainties applying a Fisher matrix method to spectroscopic galaxy clustering, weak lensing, photometric galaxy clustering, cross correlation between photometric probes, spectroscopic galaxy clustering bispectrum, CMB temperature and $E$-mode polarization, temperature-polarization cross correlation, and CMB weak lensing. We also study a nonlinear density reconstruction method to retrieve the oscillatory signals in the primordial power spectrum. We find the following percentage relative errors in the feature amplitude with $Euclid$ primary probes for the linear (logarithmic) feature model: 21% (22%) in the pessimistic settings and 18% (18%) in the optimistic settings at 68.3% confidence level (CL) using GC$_{\rm sp}$+WL+GC$_{\rm ph}$+XC. Combining all the sources of information explored expected from $Euclid$ in combination with future SO-like CMB experiment, we forecast ${\cal A}_{\rm lin} \simeq 0.010 \pm 0.001$ at 68.3% CL and ${\cal A}_{\rm log} \simeq 0.010 \pm 0.001$ for GC$_{\rm sp}$(PS rec + BS)+WL+GC$_{\rm ph}$+XC+SO-like both for the optimistic and pessimistic settings over the frequency range $(1,\,10^{2.1})$.