Kubently 2.3: one command to install, MCP everywhere, and alerts that arrive pre-diagnosed

July 2026

Kubently’s job has always been simple to say: debug Kubernetes clusters by talking to them. This release is about removing everything that stood between you and that conversation.

One command, ~75 seconds

The old quickstart was ten manual steps — namespace, three secrets, hand-written values, port-forwards. It’s now:

npm install -g @kubently/cli
kubently install

The installer creates the namespace and secrets, installs the published Helm chart (API + Redis + executor), waits for the executor to connect, and drops you straight into a debug chat. On a fresh kind cluster it clocks in at about 75 seconds, image pulls included. Reruns are idempotent.

kubently> why is my nginx pod crashlooping?

Kubently in your editor, via MCP

Kubently is now listed in the official MCP registry as io.github.kubently/kubently, and the CLI ships a stdio bridge so adding it to Claude Code is one line:

claude mcp add kubently -- kubently mcp

Your AI client gets a single natural-language tool — ask_kubently — and Kubently’s own agent does the investigation: planning, running read-only kubectl across your fleet, and returning a synthesized answer. Cursor and any other MCP client work the same way over streamable HTTP.

Proactive diagnosis: Alertmanager → Slack

The feature we’re most excited about. Point Alertmanager at Kubently’s new webhook and set a Slack incoming-webhook URL:

receivers:
  - name: kubently
    webhook_configs:
      - url: https://<your-kubently-host>/webhooks/alertmanager

Every firing alert gets diagnosed by the agent in the background, and the root-cause analysis lands in Slack — often before you’ve opened your laptop. In our testing the agent even correctly identified a synthetic always-firing alert as a canary and recommended no action. That’s the difference between an alert and an answer.

The boring-but-important parts

  • Helm chart now published at https://kubently.github.io/kubently (no more cloning the repo to install)
  • Multi-arch images (amd64 + arm64) publish automatically from main
  • api.enabled now defaults to true — a default helm install deploys the full stack
  • CLI 2.3.1 on npm with first-class unit test coverage

Get started

npm install -g @kubently/cli
kubently install

Kubently is Apache-2.0 licensed and open source on GitHub. Read-only by default, executor whitelist + RBAC enforced, works with Anthropic, OpenAI, and Google models.

Questions or feedback? Open an issue — we’d love to hear how it goes.