Recent developments in surface-enhanced Raman scattering (SERS) enable observation of single-bond vibrations in real time at room temperature. By contrast, mid-infrared (MIR) vibrational spectroscopy is limited to inefficient slow detection. Here we develop a new method for MIR sensing using SERS. This method utilizes nanoparticle-on-foil (NPoF) nanocavities supporting both visible and MIR plasmonic hotspots in the same nanogap formed by a monolayer of molecules. Molecular SERS signals from individual NPoF nanocavities are modulated in the presence of MIR photons. The strength of this modulation depends on the MIR wavelength, and is maximized at the 6-12 μm absorption bands of SiO2 or polystyrene placed under the foil. Using a single-photon lock-in detection scheme we time-resolve the rise and decay of the signal in a few 100 ns. Our observations reveal that the phonon resonances of SiO2 can trap intense MIR surface plasmons within the Reststrahlen band, tuning the visible-wavelength localized plasmons by reversibly perturbing the localized few-nm-thick water shell trapped in the nanostructure crevices. This suggests new ways to couple nanoscale bond vibrations for optomechanics, with potential to push detection limits down to single-photon and single-molecule regimes.