container-manager-multi-context

name	container-manager-multi-context
skill_type	skill
description	Operate several container backends and contexts at once — Kubernetes, Docker, Podman, and Swarm — via the container-manager-mcp MCP server's cm_multi_context tool, with per-call backend/context selection and parallel fan-out across a configured pool of contexts. Use when the agent must compare, migrate between, or simultaneously act on multiple clusters/engines/hosts. Do NOT use for single-backend operations, which are cheaper via the themed skills (container-manager-lifecycle / -swarm / -kubernetes-operations / -podman-operations).
license	MIT
tags	["container-manager-mcp","multi-context","kubernetes","docker","podman","swarm","mcp"]
metadata	{"author":"Genius","version":"0.1.0"}

Multi-Context Container Management

Domain-typed control across a pool of backends and contexts — Kubernetes clusters, Docker hosts, Podman, and Swarm managers — through the single cm_multi_context tool on the container-manager-mcp server. Each call selects a backend (and optional context) so one MCP server can drive many environments without reconnecting. Best suited to cross-cluster comparisons, staged migrations, and fan-out operations.

When to use

• Enumerate every configured backend/context and its health in one call (list_contexts).
• Run the same container/image/volume/network/service verb against a specific named context without re-authenticating per call.
• Compare state across contexts (e.g. staging vs prod Kubernetes clusters, or two Docker hosts) before/during a migration.
• Kubernetes-specific fan-out: list pods/deployments or scale a deployment in a named context + namespace.

When NOT to use

• A single, already-known backend → use the themed skill directly: container-manager-lifecycle (Docker/Podman), container-manager-swarm, container-manager-kubernetes-operations (full k8s surface — cm_multi_context's Kubernetes actions are a narrow subset: containers/images/volumes/networks/ services/pods/deployments only, not the full cm_k8s_* themed surface). container-manager-podman-operations (pods/checkpoint/kube-yaml — not exposed here).
• First-time setup of K8S_CONTEXTS/DOCKER_CONTEXTS/SWARM_CONTEXTS → see the Walkthrough below, then container-manager-config-walkthrough.

Prerequisites & environment

Connect via the mcp-client skill against the container-manager-mcp MCP server. Multi-context mode is enabled by configuring one or more context-pool env vars (or MULTI_CONTEXT_MODE=true); each pool var is a ;-separated list of name=value pairs.

Variable	Required	Notes
MULTI_CONTEXT_MODE	optional	Forces multi-context mode even with a single context configured
K8S_CONTEXTS	optional	"prod=prod-ctx;staging=staging-ctx" — kubeconfig context names
DEFAULT_K8S_CONTEXT	optional	Which K8S_CONTEXTS entry is used when context is omitted; else the first one
DOCKER_CONTEXTS	optional	"local=;remote1=<tunnel-manager-host-alias>" — Docker host aliases
DEFAULT_DOCKER_CONTEXT	optional	Default Docker context name
SWARM_CONTEXTS	optional	"cluster1=<manager-host-alias>" — Swarm manager host aliases
DEFAULT_SWARM_CONTEXT	optional	Default Swarm context name
PODMAN_ENABLED	optional	Default true; local Podman only (no multi-context Podman pool)
MULTI_CONTEXT_MAX_WORKERS	optional	Thread pool size for fan-out, default 8
MULTICONTEXTTOOL	optional	Tool toggle, default true; set false to hide cm_multi_context

Walkthrough — first run

1. Configure the pools (example: two Kubernetes clusters + one remote Docker host):

   K8S_CONTEXTS="prod=prod-cluster;staging=staging-cluster"
   DEFAULT_K8S_CONTEXT="staging"
   DOCKER_CONTEXTS="edge1=<tunnel-manager-alias>"
   MULTI_CONTEXT_MODE=true
   

2. Enumerate what's live.

   cm_multi_context action=list_contexts
   

3. Target a specific context explicitly (omit context to use the backend's default):

   cm_multi_context action=list_pods backend=kubernetes context=prod namespace=default
   

4. Proceed to the recipes below; escalate to the full themed skill (container-manager-kubernetes-operations, etc.) once you've localized the target context and need its complete tool surface.

Tools & actions

Tool	Group	Actions
cm_multi_context	Context management	list_contexts
cm_multi_context	Containers	list_containers, run_container, stop_container, remove_container, inspect_container
cm_multi_context	Images	list_images, pull_image, remove_image
cm_multi_context	Volumes	list_volumes, create_volume, remove_volume
cm_multi_context	Networks	list_networks, create_network, remove_network
cm_multi_context	Services (Kubernetes/Swarm)	list_services, create_service, remove_service
cm_multi_context	Pods (Kubernetes only)	list_pods, describe_pod
cm_multi_context	Deployments (Kubernetes only)	list_deployments, scale_deployment

Key parameters

• backend — one of kubernetes (default), docker, podman, swarm; selects which pool context is resolved against.
• context — a name from that backend's pool (e.g. a K8S_CONTEXTS key); omit to use that backend's DEFAULT_*_CONTEXT.
• name — resource name for container/image/volume/network/service verbs.
• namespace — required alongside name for describe_pod and scale_deployment (Kubernetes only); list_pods/list_deployments default to "default".
• replicas — required for scale_deployment.
• all — include stopped containers in list_containers.
• force — for stop_container/remove_container/remove_image/remove_volume/ remove_network.

Recipes

List every configured context and its resolved health:

cm_multi_context action=list_contexts

Compare pod counts across two Kubernetes clusters before a migration:

cm_multi_context action=list_pods backend=kubernetes context=staging namespace=prod
cm_multi_context action=list_pods backend=kubernetes context=prod namespace=prod

Scale a deployment in a named cluster context:

cm_multi_context action=scale_deployment backend=kubernetes context=prod name=web namespace=prod replicas=6

Pull the same image on two Docker hosts ahead of a rolling cutover:

cm_multi_context action=pull_image backend=docker context=edge1 image=myapp:2026.07.09
cm_multi_context action=pull_image backend=docker context=edge2 image=myapp:2026.07.09

Gotchas

• The Kubernetes actions here (list_pods, describe_pod, list_deployments, scale_deployment) are a narrow subset: no rollouts, no config/networking/ storage/RBAC/governance/observability themes. Once you've localized a target cluster context, switch to container-manager-kubernetes-operations (pointed at that same kubeconfig context via CONTAINER_MANAGER_KUBECONTEXT) for the full surface.
• Podman is local-only in multi-context mode — there is no PODMAN_CONTEXTS pool; backend=podman always targets the local engine.
• An unconfigured/unknown context name fails resolution — call list_contexts first to see what's actually in the pool.
• list_services/create_service/remove_service behave differently per backend: Kubernetes-shaped vs Swarm-shaped — know which backend you're targeting.
• Fan-out across many contexts runs on a bounded thread pool (MULTI_CONTEXT_MAX_WORKERS, default 8); very large pools serialize beyond that.

Related

• container-manager-config-walkthrough — how to populate K8S_CONTEXTS / DOCKER_CONTEXTS / SWARM_CONTEXTS for the first time.
• container-manager-kubernetes-operations — the full k8s tool surface once a context is localized.
• container-manager-lifecycle / container-manager-swarm / container-manager-podman-operations — the full single-backend surfaces.