Cell-free networks can effectively reduce interference due to diversity gain. Two key technologies, access point (AP) clustering and transceiver design, play key roles in cell-free networks, and they are implemented at different layers of the air interface. To address the issues and obtain global optimal results, this paper proposes an uplink joint AP clustering and receiver optimization algorithm, where a cross-layer optimization model is built based on graph neural networks (GNNs) with low computational complexity. Experimental results show that the proposed algorithm can activate fewer APs for each user with a small performance loss compared with conventional algorithms.