K8gentS — Kubernetes RCA MCP Server
Run LLM-assisted Kubernetes root-cause analysis through MCP while OPA Gatekeeper keeps the agent’s actions sandboxed.
Overview
K8gentS is an MCP server for the Operate phase that performs sandboxed, Gemini-powered root cause analysis on Kubernetes clusters.
What is this MCP server?
- Kubernetes RCA agent with MCP stdio exposure for any MCP client
- AI analysis via Google Gemini (`AI_API_KEY`, default model gemini-2.5-pro configurable with `AI_MODEL`)
- OPA Gatekeeper sandbox around the agent’s Kubernetes-facing behavior
- OCI image `docker.io/jdoornink/k8gents:0.1.1` with optional `KUBECONFIG` or in-cluster config
- OCI image docker.io/jdoornink/k8gents:0.1.1
- Default LLM model documented as gemini-2.5-pro via `AI_MODEL`
- stdio MCP transport; 3 primary env vars (AI_API_KEY required)
What problem does it solve?
When a K8s deployment fails at 2 a.m., you need fast RCA but you do not trust a raw LLM with unrestricted cluster permissions.
Who is it for?
Solo builders operating their own Kubernetes apps who want MCP-driven incident narration with policy guardrails.
Skip if: Non-Kubernetes stacks, local-only prototypes with no cluster, or teams that forbid LLM access to production credentials.
What do I get? / Deliverables
After deploying the OCI MCP server with your Gemini key and cluster access, you can invoke RCA tools from your agent with Gatekeeper-backed sandboxing.
- MCP-invokable Kubernetes RCA workflow backed by Gemini
- Policy-sandboxed agent interactions with the cluster (OPA Gatekeeper)
- Configurable model and kube context for incident sessions
Recommended MCP Servers
Journey fit
How it compares
Specialized K8s RCA MCP container—not a general cloud dashboard or passive metrics-only exporter.
Common Questions / FAQ
Who is K8gentS for?
Developers running Kubernetes in production who use MCP clients and want Gemini-assisted RCA with OPA Gatekeeper sandboxing.
When should I use K8gentS?
Use it during operate-phase error triage when pods or services misbehave and you want an agent-guided RCA path instead of solo kubectl forensics.
How do I add K8gentS to my agent?
Run the OCI package `docker.io/jdoornink/k8gents:0.1.1` as your stdio MCP server, set `AI_API_KEY` for Gemini, optionally `AI_MODEL` and `KUBECONFIG`, and register it in Claude Code or your MCP host.