Summary
Which cluster to talk to becomes an argument, so one session covers the fleet instead of one kubeconfig context at a time.
The reach comes from kubectl accepting arbitrary commands and YAML rather than from a long tool list — 3 tools cover the whole surface, and anything kubectl can express is in scope. That also sets where the boundary is: connect_cluster binds cluster-admin unless a narrower ClusterRole is named on the call, so the scoping decision is yours to make on each connection rather than one the server makes for you.
What it is
A hub-side entry point to a fleet of managed Kubernetes clusters, turning the choice of cluster into an argument on the kubectl call.
What you get
- clusters, which retrieves the Kubernetes clusters registered as managed or spoke clusters.
- connect_cluster, which generates the KUBECONFIG for a managed cluster and binds it to a ClusterRole, defaulting to cluster-admin.
- kubectl, which runs a kubectl command or applies YAML against a named cluster.
Requirements
A hub that already registers the managed clusters you want to reach, and the rights on them to generate a KUBECONFIG and bind a ClusterRole.
Setup effort
One command — npx -y multicluster-mcp-server@latest