The detection of the spread of toxic gas molecules in the air at low concentration in the field requires a robust miniaturized system combined with an analytical technique that is portable and able to detect and identify the molecules, as is the case with surface enhanced Raman scattering (SERS). This work aims to address capability gaps faced by first responders in real-time detection, identification and monitoring of neurotoxic gases by developing robust, reliable and reusable SERS microfluidic chips. Thus, the key performance attributes of a portable SERS detection system that must be addressed in detail are its limit of detection, response time and reusability. To this purpose, we integrate a 3D plasmonic architecture based on closely packed mesoporous silica (MCM48) nanospheres decorated with Au nanoparticle arrays, denoted as MCM48@Au, into a Si microfluidic chip designed and used for preconcentration and label-free detection of gases at a trace concentration level. The SERS performance of the plasmonic platform is thoroughly analyzed using DMMP as a model neurotoxic simulant over a 1 cm2 SERS active area and over a range of concentrations from 100 ppbV to 2.5 ppmV. The preconcentration-based SERS signal amplification by the mesoporous silica moieties is evaluated against dense silica counterparts, denoted as Stöber@Au. To assess the potential for applications in the field, the microfluidic SERS chip has been interrogated with a portable Raman spectrometer, evaluated with temporal and spatial resolution and subjected to several gas detection/regeneration cycles. The reusable SERS chip shows exceptional performance for the label-free monitoring of 2.5 ppmV gaseous DMMP.