Safety-gated MCP server · Proxmox VE

Proxmox, behind a safety gate.

Proxmox MCP lets an AI agent operate a Proxmox VE host — list nodes, read status, snapshot, start and stop — but every mutating call stays behind an explicit, auditable boundary instead of raw API access.

Reads open · writes gated · built on TypeMCP

What it does / 01

Full control,
never raw access.

Proxmox MCP is a boundary, not a passthrough. Reads flow freely; every mutation is named, previewable, confirmed, and logged.

01

Mutations stay behind a gate

Start, stop, snapshot, delete — every write is gated behind an explicit confirmation and a dry-run that shows what would change, so an agent can never mutate a host on a hunch.

02

A typed MCP contract

Built on TypeMCP's decorator-first typed contracts, every tool has a declared shape and a declared boundary. No hand-rolled JSON, no raw API surface leaking through.

03

Read and write, cleanly separated

Reads — nodes, VMs, containers, status — stay open and fast. Writes are a separate, named class of operation, and every one of them lands in an auditable log.

04

Works with any MCP client

Point any MCP-compatible client at it — an autonomous agent, an assistant, your own tooling. The same gated contract governs all of them the same way.

Where the line is / 02

Reads open.
Writes gated.

The whole design is one boundary drawn on purpose: what an agent can see, and what it must ask before it does.

Open source · MIT

Give an agent Proxmox, not the keys.

Get Proxmox MCP ↗

A Theorvane product · built on TypeMCP