So the kube API server runs on all control plane nodes and you can access it directly. The Rancher server also provides a proxied connection via the cattle-cluster-agent pod as well.
So if Rancher is down/unavailable you can still access your cluster by pointing your kube Config directly to one of cp nodes.
Also, you can manually generate a kubeconfig using the script https://github.com/patrick0057/kubecert in the event Rancher is down.