// Kubernetes, agentically

Debug clusters by
talking to them.

An AI agent that runs read-only kubectl across every cluster — over the A2A protocol or MCP — and tells you what's actually wrong.

Get started View on GitHub
Apache 2.0 License Python 3.13+ Kubernetes 1.28+

Speaks your agent's language

Plug Kubently into any AI client — two open protocols, same read-only cluster access.

A2A Protocol

Full conversational agent over /a2a/

Talk to Kubently in natural language. It plans, runs kubectl, and streams findings back over SSE.

curl -X POST https://kubently.io/a2a/ \
  -H "X-API-Key: $KUBENTLY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","method":"message/stream",
       "params":{"message":{"messageId":"1","role":"user",
       "parts":[{"text":"show crashing pods in prod-eu"}]}}}'

MCP New

Tools for any MCP client over /mcp

Connect Claude Code, Cursor, or your own agent. One natural-language tool — ask_kubently — puts Kubently's diagnosis loop behind your AI client. Listed in the official MCP registry.

# two lines, that's it
npm install -g @kubently/cli
claude mcp add kubently -- kubently mcp

# or connect over HTTP directly
{ "type": "streamable-http",
  "url": "https://<your-host>/mcp/",
  "headers": { "X-API-Key": "<your-api-key>" } }
Read the MCP guide →

One agent. Every cluster.

Read-only, RBAC-scoped, ~50ms round trips.

Built for real troubleshooting

Real-time debugging

~50ms command delivery over Server-Sent Events. Results stream back as they happen.

Secure by default

Read-only operations, command validation, and Kubernetes RBAC on every cluster.

AI-native

Multi-LLM support with the A2A protocol and MCP — connect any agent or provider.

Simple deployment

One API service plus lightweight executors. Deploy anywhere Kubernetes runs.

Scales horizontally

Redis pub/sub fans out commands across unlimited API pods.

Flexible integration

REST API, Node.js CLI, A2A, and MCP — meet your stack where it is.

From incident to root cause

Proactive diagnosis

Point Alertmanager at Kubently — firing alerts arrive in Slack already diagnosed.

Multi-agent systems

Full A2A implementation with tool-call interception and streaming.

Enterprise ready

OAuth/OIDC auth, TLS via cert-manager, audit logging.

Start debugging agentically

One command — kubently install — and you're chatting with your cluster in about a minute.

Get started View on GitHub