Kagent is an open-source programming framework that brings the power of agentic AI to cloud-native environments. Kagent enables AI agents to run directly in Kubernetes clusters

https://kagent.dev/docs/introduction/what-is-kagent

Helm ApplicationSet config for kagent-crds and kagent: https://github.com/olga-mir/playground/blob/main/platform/config/helm-applications/kagent-crds.yaml

kagent comes with a set of pre-built agents, armed with a number of Builtin tools. The demo includes agents and tools built on top of this framework and these manifests can be found at ai/kagent/config. These manifests are deployed as ApplicationSets:

project: ai
syncWave: "2"
name: kagent-config
path: ai/kagent/config
namespace: kagent

Example using kagent Agent kgateway-agent, which comes pre-configured with default installation.

% k explain --recursive agent.spec.tools                                                                                                                                                                                                                                                   
...
  agent <Object>
    ref <string>
  builtin       <Object>
    config      <Object>
    name        <string>
  mcpServer     <Object>
    toolNames   <[]string>
    toolServer  <string>
  type  <string>

Working example of an Agent with MCP and Agent connected (full manifest):

  tools:
  - type: McpServer
    mcpServer:
      toolServer: mcp-website-fetcher
      toolNames: ["fetch"]
  - type: Agent
    agent:
      ref: k8s-agent

But when the connected agent can't solve the task the calling agent silently ignores and does nothing: