Resources

Ark extends Kubernetes with custom resources (CRDs) in the ark.mckinsey.com API group. Each resource is reconciled by the Ark controller, so you create and manage them with standard tools — kubectl, GitOps, the Ark API, or the dashboard. This section is the field-by-field reference for every CRD; for task-oriented guides see the User Guide.

Resource reference

Resource	Kind	API version	Description
Agents	`Agent`	`v1alpha1`	Execution units that process input with a model and tools.
Models	`Model`	`v1alpha1`	LLM provider configuration and credentials.
Queries	`Query`	`v1alpha1`	A request to run work against an agent, team, model, or tool.
Teams	`Team`	`v1alpha1`	Multi-agent coordination via an execution strategy.
Tools	`Tool`	`v1alpha1`	Capabilities agents can call (HTTP, MCP, agent, team, built-in).
MCPServers	`MCPServer`	`v1alpha1`	Model Context Protocol servers that expose tools.
Memories	`Memory`	`v1alpha1`	Backing store for conversation history.
A2AServers	`A2AServer`	`v1prealpha1`	Remote Agent-to-Agent servers; discovered agents become `Agent` resources.
A2ATask	`A2ATask`	`v1alpha1`	Tracks an async task dispatched to an A2A agent.
ExecutionEngine	`ExecutionEngine`	`v1prealpha1`	Pluggable external runtime that executes agents via the A2A protocol.
ArkConfig	`ArkConfig`	`v1alpha1`	Cluster-scoped configuration singleton (e.g. default query TTL).

Every resource is namespaced except ArkConfig, which is cluster-scoped. A2AServer and ExecutionEngine are served under ark.mckinsey.com/v1prealpha1; the rest under v1alpha1.

Working with resources

All Ark CRDs behave like native Kubernetes objects:


kubectl get agents                 # list, with resource-specific print columns
kubectl describe agent my-agent    # spec, status, and events
kubectl get query my-query -o yaml # full spec + status
kubectl apply -f agent.yaml        # create or update

Each resource page documents its spec fields, status fields, and the print columns kubectl get renders.

Resource relationships

Ark resources compose into common patterns:

Agent + Model + Tools — an agent with a model and capabilities.
Team + Agents — multi-agent collaboration under a strategy.
Query + target — a request routed to an agent, team, model, or tool.
MCPServer + Tools — standardized tool integration; discovered tools become Tool resources.
Memory + Query — conversation history threaded by conversationId.
A2AServer + Agent — remote agents surfaced as local Agent resources.

See Resource Relationships for the full dependency map.