MimiOps — Kubernetes MCP server: scoped agent access
Gives an AI agent a narrow set of Kubernetes operations: read pods, logs and events, then scale or restart when something breaks. Editing manifests is off the table.
claude mcp add mimiops-mcp-server-for-kubernetes -- docker run -i --rm ghcr.io/sergelogvinov/mimiops-mcp:v0.5.0What MimiOps mcp server for Kubernetes does
The server is built around a leash. Core tools for listing and describing clusters, pods, Jobs, CronJobs, nodes, namespaces, quotas, StorageClasses, PVCs, Deployments, StatefulSets, DaemonSets, HPAs and Services are always registered, along with pod logs and event lookups. Anything that mutates state — pod or Job deletion, CronJob suspend and resume, workload scaling — sits behind --allow-destructive, which is off unless you turn it on.
Extensions arrive through --extensions. Helm is loaded by default and covers release listing and status, with rollback gated the same way. FluxCD is opt-in: read GitRepository, OCIRepository, HelmRelease and Kustomization objects, and with destructive mode also trigger reconciliation or suspend and resume resources. Karpenter support lists NodePools with node counts and CPU and memory headroom.
Point it at a cluster with --kubeconfig or the KUBECONFIG variable; --context and --namespace narrow the scope. You can run it from the published Docker image ghcr.io/sergelogvinov/mimiops-mcp:v0.5.0 or install the binary with Homebrew, then wire it into a client via JSON config or an MCPB bundle. Arbitrary YAML apply and secret changes are blocked by design, so manifest edits stay outside the agent's reach.
What it helps with
- find out why a pod keeps restarting
- read Job logs to locate a failure
- check a Helm release status and roll it back
- see which FluxCD object owns a resource
- check NodePool capacity headroom
Example prompts for the agent
- Show me the logs of pod
api-7d9fin namespaceprodand tell me why it fails - List Helm releases in the cluster and the status of the last one
- Scale the
workerDeployment to two replicas
How to install MimiOps mcp server for Kubernetes
Platform CLI
claude mcp add mimiops-mcp-server-for-kubernetes -- docker run -i --rm ghcr.io/sergelogvinov/mimiops-mcp:v0.5.0Platform CLI
codex mcp add mimiops-mcp-server-for-kubernetes -- docker run -i --rm ghcr.io/sergelogvinov/mimiops-mcp:v0.5.0Config file
{
"mcpServers": {
"mimiops-mcp-server-for-kubernetes": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/sergelogvinov/mimiops-mcp:v0.5.0"
]
}
}
}Config file
{
"servers": {
"mimiops-mcp-server-for-kubernetes": {
"type": "stdio",
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/sergelogvinov/mimiops-mcp:v0.5.0"
]
}
}
}Platform CLI
gemini mcp add mimiops-mcp-server-for-kubernetes docker run -i --rm ghcr.io/sergelogvinov/mimiops-mcp:v0.5.0Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about MimiOps mcp server for Kubernetes
How do I install MimiOps mcp server for Kubernetes in Claude Code?
Run claude mcp add mimiops-mcp-server-for-kubernetes -- docker run -i --rm ghcr.io/sergelogvinov/mimiops-mcp:v0.5.0 and restart the client afterwards.
What does MimiOps do?
It exposes a fixed set of Kubernetes tools to an agent: cluster state, logs and events, plus Helm, FluxCD and Karpenter inspection. Mutating calls stay behind --allow-destructive.
What does the server need to connect?
A kubeconfig, passed as --kubeconfig or through KUBECONFIG. Without it the server cannot reach the cluster; in Zed the file path goes into the env block under context_servers.