Apriori: The openshift/openshift-mcp-server (downstream) and containers/kubernetes-mcp-server (upstream) are already able to integrate with the kube-apiserver running the OpenShift cluster. The current guidance is that we have one MCP server per API and not develop individual MCP servers per component, as the main benefit will be not having to re-implement common features such as read-only mode, authentication or redaction, etc.

Existing Proposal for routing based MCP: https://docs.google.com/document/d/1L0yyMiOnrltxH_iN0OztHlmK3YCC9smhuQ-_QcI77E8/edit?tab=t.0 

Current Plan: Within the openshift/k8s MCP server come up with team-specific system prompts that can be dynamically injected based on the type of query. 

Acceptance criteria: a detailed prompt with examples that works {validated across atleast few set of differently sized models (i.e. small eg. Granite, medium eg. Gemini Flash, large eg. Claude Sonnet, Gemini Pro, etc.)} to allow the users ability to collect a must-gather from an openshift cluster.