Background Pencil Beam Scanning proton therapy has many advantages from a therapeutic point of view, but raises technical constraints in terms of treatment verification. The treatment relies on a large number of planned pencil beams (PB) (up to thousands), whose delivery is divided in several low‐intensity pulses delivered a high frequency (1 kHz in this study). Purpose The purpose of this study was to develop a three‐dimensional quality assurance system allowing to verify all the PBs’ characteristics (position, energy, intensity in terms of delivered monitor unit—MU) of patient treatment plans on a pulse‐by‐pulse or a PB‐by‐PB basis. Methods A system named SCICOPRO has been developed. It is based on a 10 × 10 × 10 cm$^3$ scintillator cube and a fast camera, synchronized with beam delivery, recording two views (direct and using a mirror) of the scintillation distribution generated by the pulses. A specific calibration and analysis process allowed to extract the characteristics of all the pulses delivered during the treatment, and consequently of all the PBs. The system uncertainties, defined here as average value + standard deviation, were characterized with a customized irradiation plan at different PB intensities (0.02, 0.1, and 1 MU) and with two patient's treatment plans of three beams each. The system's ability to detect potential treatment delivery problems, such as positioning errors of the treatment table in this work (1° rotations and a 2 mm translation), was assessed by calculating the confidence intervals (CI) for the different characteristics and evaluating the proportion of PBs within these intervals. Results The performances of SCICOPRO were evaluated on a pulse‐by‐pulse basis. They showed a very good signal‐to‐noise ratio for all the pulse intensities (between 2 × 10$^{−3}$ MU and 150 × 10$^{-3}$ MU) allowing uncertainties smaller than 580 µm for the position, 180 keV for the energy and 3% for the intensity on patients treatment plans. The position and energy uncertainties were found to be little dependent from the pulse intensities whereas the intensity uncertainty depends on the pulses number and intensity distribution. Finally, treatment plans evaluations showed that 98% of the PBs were within the CIs with a nominal positioning against 83% or less with the table positioning errors, thus proving the ability of SCICOPRO to detect this kind of errors. Conclusion The high acquisition rate and the very high sensitivity of the system developed in this work allowed to record pulses of intensities as low as 2 × 10$^{−3}$ MU. SCICOPRO was thus able to measure all the characteristics of the spots of a treatment (position, energy, intensity) in a single measurement, making it possible to verify their compliance with the treatment plan. SCICOPRO thus proved to be a fast and accurate tool that would be useful for patient‐specific quality assurance (PSQA) on a pulse‐by‐pulse or PB‐by‐PB verification basis.