KubeStellar MCP: AI-Native Multi-Cluster Management for Claude Code
June 2026
If you use Claude Code to manage Kubernetes, you’ve always had to drop out of your AI session to run kubectl commands against multiple clusters. Copying context names, switching configs, translating single-cluster commands into multi-cluster workflows — it breaks flow.
kubestellar-mcp fixes that.
What It Is
kubestellar-mcp is a set of Model Context Protocol (MCP) tools that gives Claude Code (and any MCP-compatible AI client) native multi-cluster Kubernetes capabilities. Instead of being a single-cluster tool that happens to be called from AI, it’s designed from the ground up to let you work with your applications — not your clusters.
Two binaries, each with a focused purpose:
| Tool | What It Does |
|---|---|
| kubestellar-ops | Multi-cluster diagnostics, RBAC analysis, security posture checks across all clusters at |
| kubestellar-deploy | App-centric deployment, GitOps workflows, intelligent workload placement using KubeStellar policies |
Install in Under a Minute
brew tap kubestellar/tap
brew install kubestellar-ops kubestellar-deploy
``` installed, Claude Code picks them up automatically on next launch.
---
## What You Can Do
With `kubestellar-ops` active in your Claude Code session:
```text
"Check RBAC permissions for the payments service across all my clusters"
"Which clusters have pods in CrashLoopBackOff right now?"
"Show me the security posture diff between staging and production"With kubestellar-deploy:
"Deploy the payments service to all clusters in us-east with label env=prod"
"Set up GitOps sync for this repo across my fleet"
"Place the ML inference workload on clusters with GPU nodes"The tools understand KubeStellar’s multi-cluster placement model natively — you describe intent, they handle distribution.
Why MCP for Multi-Cluster?
Single-cluster kubectl works fine for cluster. At 5, 10, or 50 clusters, the cognitive load of managing contexts, aggregating state, and reasoning about placement becomes the bottleneck — not the work itself.
MCP tools run in the same context as your AI session. Claude can call kubestellar-ops to fetch cluster state, reason about it, and call kubestellar-deploy to act — all within a single conversation. The multi-cluster complexity is handled by the tools; you stay focused on the outcome.
Try It and Tell Us What to Build Next
kubestellar-mcp is early. The tools work, but there’s a lot more to build — better placement reasoning, Fleet visibility, policy recommendations, event streaming.
Try it, break it, and open an issue or join the conversation on Slack with what you need.
Full Documentation
→ kubestellar-mcp documentation
The KubeStellar Team