We present a real-time LiDAR-Inertial-Camera SLAM system with 3D Gaussian Splatting as the mapping backend. Leveraging robust pose estimates from our LiDAR-Inertial-Camera odometry, Coco-LIC, an incremental photo-realistic mapping system is proposed in this paper. We initialize 3D Gaussians from colorized LiDAR points and optimize them using differentiable rendering powered by 3D Gaussian Splatting. Meticulously designed strategies are employed to incrementally expand the Gaussian map and adaptively control its density, ensuring high-quality mapping with real-time capability. Experiments conducted in diverse scenarios demonstrate the superior performance of our method compared to existing radiance-field-based SLAM systems.