Znote Recipes

Docker and Kubernetes — living runbooks

Container status, pod diagnosis, deployment drift — ops notes that check the real infrastructure state.

MCP · Docker and Kubernetes — infrastructure state, inline

Use case: living runbooks — "what's running?", "why does this pod keep restarting?" — the answer inside the ops note, next to the procedure.

Configuration (.znote/mcp.json)

{
  "mcpServers": {
    "docker": {
      "command": "uvx",
      "args": ["mcp-server-docker"]
    },
    "kubernetes": {
      "command": "npx",
      "args": ["-y", "mcp-server-kubernetes"],
      "env": { "ALLOW_ONLY_NON_DESTRUCTIVE_TOOLS": "true" }
    }
  }
}

mcp-server-docker talks to the local Docker daemon; mcp-server-kubernetes (Flux159) uses your current kubeconfig. The ALLOW_ONLY_NON_DESTRUCTIVE_TOOLS flag removes delete/exec — recommended.

Exposed tools: Docker → list_containers, fetch_container_logs, create_container… · K8s → kubectl_get, kubectl_describe, kubectl_logs, kubectl_rollout…

Examples

Local diagnosis

List the running containers with their uptime and usage. Does the
"postgres-dev" container show errors in its last 50 log lines?

Namespace health

In the "production" namespace: which pods are not Running or restarted
in the last 24h? For each, the probable reason from describe and the
latest logs.

Living runbook

Check the "api" deployment: ready replicas, image version, last
rollout. Compare with the procedure in @runbooks/deploy-api.md and
flag any drift.

Tips

A runbook note = the procedure in markdown plus the blocks that check the real state: documentation and diagnosis in one place.
Keep the non-destructive mode; actions (restart, scale) deserve a human in the loop.