Endoscopy provides an intuitive observation of the internal organ structures for surgeons. Most clinical 3D endoscopes generally rely on visual parallax to create a stereoscopic vision, which is incapable of obtaining quantitative depth information for precise diagnosis. Hence a compact probe-type 3D quantitative imaging remains a challenging task for the endoscopic system. In this work, we propose an integrated light-field endoscope that realizes 3D morphology observation with cellular-level resolution from a single-frame data record. In our prototype, a compact customized GRIN lens array (GLA) is used as an objective lens to capture the spatio-angular information directly from object space, as well as maintain consistent spatial measurements. In data processing, we employ a calibration method to estimate the 3D point spread function. After decoupling the spatio-angular information, the 3D morphology of sample can be reconstructed by a deconvolution operation. The system has a lateral resolution of about 15∼40 μm and an axial resolution of 100∼200 μm over an axial operating range of 10 mm. In experiment, we demonstrate the 3D imaging capacity of our platform via fixed structures and living worms.