# Proximo

An MCP server that lets an AI agent operate a Proxmox virtualization cluster by planning every risky action first, taking a snapshot where possible, and writing a signed receipt of what changed.

Proximo answers a plain-English request such as diagnosing why a container is thrashing by pulling node and guest status, tailing logs and running a diagnostic inside the container. Any change is read-only by default: the first call for a risky action returns its blast radius as a plan rather than applying it, a snapshot is taken first when the operator asks and the action refuses to run if it cannot snapshot, and a signed, tamper-evident receipt records exactly what changed afterward.

It runs against a user's own Proxmox cluster and token, covering all four Proxmox products, PVE, PBS, PMG and PDM, with no external service in the loop, and every dispatched call passes through one governed trust layer that is itself bounded by the permissions Proxmox already enforces. A verification guide walks through independently checking that a tampered receipt is rejected and that no outbound network traffic leaves the tool.

Website: https://john-broadway.github.io/proximo/