Proxmox AIops

Disclaimer: Community-maintained open-source project. Not affiliated with, endorsed by, or sponsored by Proxmox Server Solutions GmbH. "Proxmox" is a trademark of its owner. MIT licensed.

AI-powered Proxmox VE VM and container lifecycle operations with a built-in governance harness — unified audit log, token/runaway budget guard, undo-token recording, and descriptive risk-tier labels. Self-contained: no external dependencies beyond proxmoxer and the MCP SDK. Coverage is not yet exhaustive across every Proxmox operation.

Verification status: live-verified against real Proxmox VE 8.4.19 across three rounds — read-only surfaces, the QEMU write surface, and a two-node cluster (quorum, live migration, move-disk, a backup that actually succeeded). Rounds 2 and 3 each found a real bug the mocks could not see. See docs/VERIFICATION.md for exactly what was proven and what is still uncovered.

What works

• CLI (proxmox-aiops ...): vm list/get/config/start/stop/shutdown/reboot/reconfigure/clone/delete/migrate, vm resize-disk/move-disk/agent-ping, vm snapshot-create/snapshot-delete/snapshot-list/snapshot-rollback, backup create/list/restore, ct list/start/stop, cluster nodes/status/task-status/resources/node-status/task-log/next-vmid, ha status/resources, pool list/members, firewall vm-rules/cluster-status, storage list/content, diagnose node-pressure/guest-health, undo list/apply, init, secret set/list/rm/migrate/rotate-password, doctor, mcp.
• MCP server (proxmox-aiops mcp or proxmox-aiops-mcp): 43 tools, every one wrapped with the bundled @governed_tool harness.
• Diagnostics / RCA (read-only): diagnose node-pressure ranks cluster nodes by CPU/memory/root-fs pressure; diagnose guest-health scans VMs/containers for stopped guests, memory saturation, and disks near full. Every finding cites the measured number that tripped it and a concrete action — transparent heuristics, not a black-box verdict.
• Credentials: proxmox-aiops init (onboarding wizard) and proxmox-aiops secret ... manage an encrypted secret store — no plaintext passwords in config.yaml.
• Reversibility: write ops with a clean inverse (start/stop/shutdown/reconfigure/clone/migrate/snapshot-create/move-disk, container start/stop, and restore-into-a-free-vmid) record an inverse undo descriptor; irreversible ops (delete, snapshot-rollback, forced restore) declare none and are tagged high risk. Disk resize is grow-only (shrink refused).
• Async tasks: Proxmox writes return a task UPID — poll completion with cluster task-status / read lines with cluster task-log (the runaway budget guard prevents poll loops from running away).

What this tool does, and does not, decide

It delivers Proxmox VE operations — reads and writes — accurately and efficiently, and records every one of them. It does not decide whether a write is allowed to happen. That is the agent's judgement, or the permission of the account you connect it with: use a Proxmox VE user or API token granted only read privileges (no VM./Datastore. write roles), and the writes fail at the server — the place that actually owns the permission.

So there is no read-only switch, no policy file, no approval gate to configure. The one thing the tool guarantees is that nothing is silent: every call, over MCP and over the CLI alike, lands an audit row in ~/.proxmox-aiops/audit.db, and destructive writes still capture their before-state and record an inverse where one exists.

Each tool declares a risk_level, carried into the audit row as a descriptive tier (none/confirm/review) — so a reviewer can see at a glance that a row was a high-risk delete. It is a label, not a gate.

Running a smaller / local model? See agent-guardrails.md — it lists the guardrails this tool now enforces for you (so you don't spend prompt budget restating them) and gives a ready-made system prompt for what's left.

Quick start

As a Claude Code plugin

One install gives an agent both the skill and the MCP server:

/plugin marketplace add AIops-tools/marketplace
/plugin install proxmox-aiops@aiops-tools

The MCP server is fetched with uv and pinned to the package version this plugin declares, so an audit row can be traced back to the code that wrote it. Credentials are still configured with proxmox-aiops init — see below.

As an OpenClaw plugin

The same bundle is published on ClawHub, where one install delivers the skill and its MCP server together:

openclaw plugins install clawhub:@zw008/proxmox-aiops
openclaw skills info proxmox-aiops          # expect: Visible to model: yes

Restart the OpenClaw gateway afterwards so it loads the plugin. The MCP server is fetched with uv, pinned to this exact release, so uvx has to be on PATH — without it the skill still installs but reports Visible to model: no. Credentials are configured exactly as below.

As a CLI or standalone MCP server

uv tool install proxmox-aiops
mkdir -p ~/.proxmox-aiops
# create ~/.proxmox-aiops/config.yaml with a targets: list
# put secrets in ~/.proxmox-aiops/.env  (chmod 600)
proxmox-aiops doctor

Example ~/.proxmox-aiops/config.yaml:

targets:
  - name: pve-lab
    host: 10.0.0.10
    user: "claude@pve!ops"    # API token: user@realm!tokenid — a dedicated,
                              # least-privilege user, NOT root@pam
    node: pve1
    auth_kind: token
    verify_ssl: false          # self-signed lab certs only

~/.proxmox-aiops/.env (chmod 600): PROXMOX_PVE_LAB_SECRET=<token-uuid>

Audit & safety

All operations are logged to a local SQLite audit DB under ~/.proxmox-aiops/ (relocatable via PROXMOX_AIOPS_HOME). Every write tool passes through the governance harness: token/runaway budget guard, risk-tier tagging, and audit logging. Destructive CLI commands (vm stop, vm delete, vm snapshot-delete, vm snapshot-rollback, ct stop) require double confirmation and support --dry-run (notably backup restore, which is high risk). API-returned text is run through a prompt-injection sanitizer.

Contributing & feature requests

Coverage is intentionally focused. Missing a device, action, or feature you need? Open an issue or pull request at github.com/AIops-tools/Proxmox-AIops — feature requests, contributions, and comments are all welcome.

License: MIT.