The existing solutions for multi-user detection in uplink (UL) grant-free sparse code multiple access (SCMA) rely on the prior knowledge of user sparsity, i.e., the number of active users. An alternative solution, which sets the sparsity as a statistically empirical value to get a rough active user set and then eliminates the false detected inactive users with joint message passing algorithm (JMPA), leads to either increasing computation complexity of JMPA or high missed detection probability. In this paper, we propose a receiver for UL grant-free SCMA which relies on no prior knowledge of user sparsity. We propose a detection-based group orthogonal matching pursuit (DGOMP) active user detector to get an accurate active user set rather than a rough active user set. Then we modify the JMPA by taking the channel gain and noise power into consideration when calculating the prior information of the zero codeword. The modified JMPA helps to further eliminate the false detections caused by noise, channel fading and non-orthogonality of pilot sequences. Simulation results show that our proposed receiver without prior knowledge of user sparsity has acceptable performance degradation compared with currently existing solution with ideal, however unable to get in practice, prior knowledge of user sparsity.