| name | container-manager-podman-operations |
| skill_type | skill |
| description | Rootless Podman pod/kube operations via the container-manager-mcp MCP server — pods (create/list/stats/top/inspect/logs/stop/rm), Kubernetes YAML interop (generate/play kube), checkpoint/restore, pod-scoped networks and volumes, health checks, and system prune. Use when the agent must drive Podman pod-level workloads beyond single containers. Do NOT use for single-container Docker/Podman lifecycle (container-manager-lifecycle) or for a real Kubernetes cluster (container-manager-kubernetes-operations). |
| license | MIT |
| tags | ["container-manager-mcp","podman","pods","rootless","checkpoint","mcp"] |
| metadata | {"author":"Genius","version":"0.1.0"} |
Podman Pod/Kube Operations
Domain-typed control of Podman pods and Podman-specific features through
the single cm_podman tool on the container-manager-mcp server. Podman
groups containers into local pods (its rootless analogue of a Kubernetes pod) and
adds checkpoint/restore and Kubernetes-YAML interop that plain Docker doesn't have.
Prefer this cm_* tool over raw podman shell.
When to use
- Create/list/inspect/stop/remove a Podman pod, and read pod-scoped stats, top,
or logs.
- Generate a Kubernetes YAML manifest from a running pod (
podman_generate_kube_yaml)
or play a YAML manifest into local pods (podman_play_kube_yaml) — useful for
Kubernetes-migration dry runs on a single host before targeting a real cluster.
- Checkpoint a running container to disk and restore it later (live-migration /
fast-restart workflows unique to Podman's CRIU integration).
- Manage pod-scoped networks and volumes, run a container health check, or prune
the Podman system.
When NOT to use
- Single-container list/logs/stop/remove/exec, or plain images/compose →
container-manager-lifecycle (works for Podman too — it auto-detects the engine).
- A real Kubernetes cluster (not just YAML interop) →
container-manager-kubernetes-operations.
- Docker Swarm →
container-manager-swarm.
- Operating Podman alongside other backends/contexts at once →
container-manager-multi-context.
Prerequisites & environment
Connect via the mcp-client skill against the container-manager-mcp MCP server.
A reachable Podman engine is required (rootless or rootful).
| Variable | Required | Notes |
|---|
CONTAINER_MANAGER_TYPE | optional | Set to podman to force Podman (else auto-detect) |
PODMANTOOL | optional | Tool toggle, default true; set false to hide cm_podman |
Tools & actions
| Tool | Group | Actions |
|---|
cm_podman | Kubernetes interop | podman_generate_kube_yaml, podman_play_kube_yaml |
cm_podman | Checkpoint/restore | podman_checkpoint, podman_restore |
cm_podman | Pod management | podman_pod_create, podman_pod_list, podman_pod_stats, podman_pod_top, podman_pod_inspect, podman_pod_logs, podman_pod_stop, podman_pod_rm |
cm_podman | Network | podman_network_create, podman_network_list, podman_network_inspect |
cm_podman | Volume | podman_volume_create, podman_volume_list, podman_volume_inspect |
cm_podman | System | podman_system_prune, podman_health_check |
Key parameters
pod_name — required for every pod-scoped action (podman_pod_*) and for
podman_generate_kube_yaml.image + command — required for podman_pod_create (the pod's infra + first
container).yaml_path — required for podman_play_kube_yaml.namespace — Kubernetes namespace metadata used by podman_generate_kube_yaml
(default "default"); Podman itself is not namespace-scoped.container_id + checkpoint_dir — required for both podman_checkpoint and
podman_restore.network_name / driver (default "bridge") / subnet — for
podman_network_create.volume_name / driver (default "local") — for podman_volume_create.tail_lines (default 100) — for podman_pod_logs.config — a dict health-check spec, required with container_id for
podman_health_check.
Walkthrough — first run
- Confirm the engine is Podman.
cm_info_operations action=get_info
- Create a pod and inspect it.
cm_podman action=podman_pod_create pod_name=web-pod image=nginx:latest
cm_podman action=podman_pod_list
cm_podman action=podman_pod_inspect pod_name=web-pod
- Verify health and logs.
cm_podman action=podman_health_check container_id=<id> config={"test":["CMD","curl","-f","http://localhost/"]}
cm_podman action=podman_pod_logs pod_name=web-pod tail_lines=100
Recipes
Migration dry run — snapshot a running pod to Kubernetes YAML, then replay it:
cm_podman action=podman_generate_kube_yaml pod_name=web-pod namespace=prod
cm_podman action=podman_play_kube_yaml yaml_path=/srv/web-pod.yaml
(Once validated, hand the same manifest to a real cluster via
container-manager-kubernetes-operations.)
Checkpoint a container before a host maintenance window, restore after:
cm_podman action=podman_checkpoint container_id=<id> checkpoint_dir=/srv/checkpoints/web-2026-07-09
# ... host maintenance ...
cm_podman action=podman_restore container_id=<id> checkpoint_dir=/srv/checkpoints/web-2026-07-09
Create an isolated pod network with an explicit subnet:
cm_podman action=podman_network_create network_name=web-net driver=bridge subnet=10.89.10.0/24
Sustain: prune unused pods/images/volumes on a schedule:
cm_podman action=podman_system_prune
Gotchas
- Pod-scoped actions (
podman_pod_*) act on the pod, not individual containers
inside it — use container-manager-lifecycle's cm_container_operations for a
single container's logs/exec once you have its id from podman_pod_inspect.
podman_checkpoint/podman_restore require CRIU support on the host; unsupported
kernels/containers fail — treat this as an availability check, not a guarantee.podman_play_kube_yaml is a local Podman-pod replay, not a real cluster
apply — it doesn't create Kubernetes API objects; use
container-manager-kubernetes-operations for that.podman_system_prune is destructive and unbounded — confirm intent before running.config for podman_health_check is a raw dict (not a JSON string, unlike some
cm_k8s_config fields) — pass it as a native object.
Related
container-manager-lifecycle — single-container Docker/Podman verbs (also
works against a Podman engine).container-manager-kubernetes-operations — the real Kubernetes cluster surface
that podman_generate_kube_yaml/podman_play_kube_yaml manifests target next.container-manager-multi-context — operate Podman alongside other backends.
