Verify the Kubernetes Agent

Verify the installation from IncidentFlow, from Kubernetes, and from an MCP client.

Get a verification checklist in MCP

Ask: “Verify my Kubernetes Agent connection.” IncidentFlow MCP calls the structured integration_guide with integration="kubernetes" and goal="verify" only after Helm installation is complete, or when you explicitly ask to verify. Pass the same cluster_identifier used at registration. It gives the rollout check, connection-health check, recent heartbeat expectation, and the read-only RBAC check. It must call k8s_agent_status for that exact identifier before reporting healthy: true. Use live Kubernetes tools for current cluster facts; do not use workspace knowledge search as a substitute for a live connection check.

Check the IncidentFlow application

Open IncidentFlow → Kubernetes Agents and select the cluster. The expected state is:

• connection status: Connected;
• agent status: Online;
• cluster identifier matches the installed cluster;
• the latest heartbeat is recent;
• the reported agent version is the expected version.

Understand setup checks

The setup page checks, in order: registration token accepted, agent credentials issued, gateway connected, heartbeat received, and read-only evidence permissions verified. The page polls every five seconds. A first heartbeat normally arrives within two minutes; if it does not, inspect the Helm release and see Troubleshooting.

Check the Kubernetes workload

Confirm that the agent deployment is available:

kubectl get deployment,pods \
  --namespace incidentflow-agent

Inspect the rollout:

kubectl rollout status deployment/incidentflow-k8s-agent \
  --namespace incidentflow-agent

The deployment should complete successfully and the agent pod should be Running and Ready.

Check recent agent logs

kubectl logs \
  --namespace incidentflow-agent \
  deployment/incidentflow-k8s-agent \
  --tail=100

The logs should show a successful registration or gateway connection. They should not contain authentication loops, permission errors, or repeated connection failures.

Verify through MCP

Call k8s_agent_status with the exact cluster_identifier. A healthy response contains:

{
"status": "connected",
"healthy": true
}

Then call k8s_connection_health to verify the end-to-end connection, and k8s_list_namespaces to confirm the expected read-only visibility.

Verification checklist

• The deployment rollout is complete.
• The agent pod is ready.
• IncidentFlow shows the agent as connected.
• The latest heartbeat is recent.
• The reported cluster and agent version are correct.
• MCP Kubernetes tools return the expected cluster context.
• The agent has no Kubernetes write permissions.