wtf is rocketplaneio?

rocketplaneIO is a self hosted platform, written in Go, that watches over a Kubernetes cluster and lets an AI agent both diagnose and fix problems it finds. The observability side uses a technology called eBPF to capture traces, logs, and metrics from every service in your cluster automatically, without you having to add any tracking code or extra libraries to your applications. It can follow a request as it moves between services, including ones built as compiled Go binaries, and shows the resulting failures and error logs together on one connected map of your services. What sets it apart is the second half: rather than only telling you something is wrong, it exposes an interface, using a standard called MCP, that lets an AI coding assistant such as Claude Code or Cursor connect, read all of that observability data, and then take real actions on the cluster, such as reading or changing Kubernetes resources. Every action the agent proposes is wrapped in a transaction that saves a snapshot of the previous state first, so any change can be rolled back afterward. Actions are sorted into four risk levels, from safe read only operations to destructive ones like deletions or scaling something down to zero, and anything disruptive or destructive is paused until a human approves it in the user interface, with a full audit trail showing exactly what the agent read, changed, and who approved it. To get started, you need Docker and a Kubernetes cluster such as minikube, and a single install script sets up the whole platform along with a web interface. Connecting your own cluster is done through a generated command that installs a lightweight agent and the eBPF collector, and connecting an AI assistant is done by creating an API key and adding rocketplaneIO as an MCP server. The project is currently in an early alpha stage, so its interfaces may still change, and it is not recommended for production use yet. It is released under the Apache 2.0 license and has 225 stars.