The proposed STCF (Super Tau Charm Facility) is an electron-positron collider with high luminosity. Under the conditions of a high radiation, high counting-rate environment, and a pure physics event rate of up to 400 kHz, for a Level 1 (L1) trigger system, it is crucial to suppress background events to an acceptable rate. To this end, this study presents a three-dimensional (3D) track reconstruction algorithm for the pre-research of the STCF main drift chamber (MDC) L1 trigger. The proposed algorithm can reject tracks outside the interaction region and provide 3D track information to the global trigger logic (GTL) of an L1 trigger for further analysis with other sub-detectors. By using hit information from the MDC and the transverse momentum (pT ) and azimuthal angle (ϕ) from previous research on two-dimensional reconstruction, this study reconstructs the z (longitudinal) position for the track vertex using a neural network-based method with an innovative stereo track segment design. The neural network is trained using high granularity quantization (HGQ) to reduce resource consumption while maintaining resolution. The proposed method is implemented into a field programmable gate array (FPGA) using the hls4ml. The achieved resolution of the z-vertex reconstruction of a single track (z ∈ [-50, 50]cm) is approximately 2.8 cm, which can ensure rejecting 97% beam background tracks in a ± 3σ interval.