trackmcp
Back to directory
ahmetem

proxmox-mcp-suite

View on GitHub

Proxmox MCP Suite

3 starsOthers Updated Aug 10, 2026
claude-codehomelabmcpmodel-context-protocolproxmoxproxmox-backup-serverproxmox-vevirtualization

Documentation

Proxmox MCP Suite

Run your Proxmox VE cluster and Backup Server from Claude — deeply, safely, and without drowning the context window.

·

·

·

·


Two Model Context Protocol servers, packaged

as Claude Code plugins, that turn Claude into a capable Proxmox operator

69 tools across virtualization and backup, engineered to run *inside an

LLM's context*: compact output, one-call health, inline task results, and a

two-tier safety model with dry-run previews and a tamper-evident audit trail.

⚡ Quick start

Prereq: Claude Code and `uv` (`uvx`). The servers run via `uvx` — nothing to `pip install` by hand.

text
/plugin marketplace add ahmetem/proxmox-mcp-suite
/plugin install proxmox-ve@proxmox-mcp-suite
/plugin install proxmox-backup@proxmox-mcp-suite

Export your Proxmox credentials (see Configuration), restart

Claude Code, and ask:

> *"How is the server? Any storage filling up?"* — one `proxmox_health_overview` call.

>

> *"Snapshot VM 102 as pre-upgrade, then reboot it."*

>

> *"Is my last backup of CT 200 healthy?"*

🎯 Why this suite

Most Proxmox MCP servers wrap the API and hand Claude raw JSON. This one is

built for the thing that actually constrains an agent — context — and for

not breaking your infrastructure.

Proxmox MCP SuiteTypical Proxmox MCP
OutputCompact, length-capped JSON · one-call `health_overview` · list filters · `fields` projectionVerbose full-object dumps
Task launches`wait_seconds` polls the task and returns the final result inlineFire, then a second call to check status
MutationsTwo-tier gate: `confirm` + `i_understand_data_loss`, plus `dry_run` previews on the risky onesSingle flag, or nothing
AuditabilityHost/guest shell exec appended to a hash-chained, tamper-evident log (`proxmox_audit_verify`)Plain log, or none
Depth (VE)Guests, snapshots, backups, storage, disks, LVM, full ZFS, disk-prep, provisioning, task forensics, backup self-healVM/CT lifecycle only
Backup (PBS)Dedicated server: datastores, snapshots, GC, verify, prune, read-only by defaultUsually absent
InstallOne marketplace, `uvx` — zero manual Python setupClone + venv + pip + JSON config

> Modular by design — install only Proxmox VE, only PBS, or both.

📦 What's inside

`proxmox-ve` — 52 tools

Repo: `homelab-proxmox-mcp`

  • Guests — list/status, create VM & LXC from scratch, clone, power, resize
  • Snapshots — create / rollback / delete (data-loss gated)
  • Backups — vzdump create / list / restore (refuses silent overwrite)
  • Storage & disks — pools, usage breakdown, physical disks + SMART
  • LVM / ZFS — VG/thin, full ZFS: pools, datasets, snapshots, scrub, send, properties, disk-prepare & provisioning
  • Exec — guest-VM SSH, host SSH (free + a read-only allow-listed variant safe for agents), LXC `pct exec`, service control, log tail
  • Forensics & self-heal — task list/logs, backup-job inspection, stale `@vzdump` snapshot cleanup that unblocks failing backups

`proxmox-backup` — 17 tools

Repo: `homelab-pbs-mcp`

  • Datastores & snapshots — status, usage, list, protect/forget
  • Maintenance — garbage collection, verify jobs, prune (with dry-run)
  • Tasks — status and logs by UPID
  • Read-only by default — write-side tools stay inert until `PBS_ALLOW_WRITE=true`

🛡️ Safety model

Read-only tools run freely. Everything that changes state requires `confirm=true`;

anything that destroys persistent data *additionally* requires

`i_understand_data_loss=true` — so Claude only fires these after you clearly ask.

  • `dry_run` previews on the highest-consequence mutations (create VM/CT,

clone, restore) return the exact API call they *would* make (secrets masked),

without touching anything.

  • Tamper-evident audit — every host/guest shell exec is appended to a

hash-chained log (SHA-256, or HMAC-SHA256 with `PROXMOX_AUDIT_HMAC_KEY`).

`proxmox_audit_verify` recomputes the chain and flags any altered, deleted,

or reordered line.

  • Token auth, never root passwords — API tokens are revocable and scope the

blast radius.

⚙️ Configuration

Credentials are passed to the servers via environment variables — export them in

the shell that launches Claude Code (no secret is stored in the plugin).

Proxmox VE — required: `PROXMOX_HOST`, `PROXMOX_USER`, `PROXMOX_TOKEN_NAME`,

`PROXMOX_TOKEN_VALUE`. Optional: `PROXMOX_PORT`, `PROXMOX_VERIFY_SSL`, and the

SSH-backed tools' `PROXMOX_SSH_*`. → full list in the

server repo.

Proxmox Backup Server — required: `PBS_HOST`, `PBS_TOKEN_ID`,

`PBS_TOKEN_SECRET`, `PBS_NODE`. Optional: `PBS_VERIFY_TLS`,

`PBS_DEFAULT_DATASTORE`, and `PBS_ALLOW_WRITE` (gates GC/prune/forget). →

PBS server repo.

Verify servers loaded with `/mcp`.

🔧 How it runs

Each plugin launches its server with `uvx`, which fetches, installs and isolates

it automatically from PyPI (`pve-mcp`,

`pbs-mcp`) — no clone, no virtualenv, no

manual `pip install`. `uvx` caches the environment after the first launch, so

later starts are instant. To pin a specific release, set a plugin's `args` to

e.g. `["pve-mcp==1.2.0"]`.

🔒 Security notes

  • Keep the Proxmox / PBS APIs on a trusted LAN or behind a VPN.
  • Privilege-separate the API tokens; grant the narrowest role that works.
  • Don't remove the `confirm` / `i_understand_data_loss` guards.

📄 License

GNU General Public License v3.0.

Not affiliated with or endorsed by Proxmox Server Solutions GmbH. "Proxmox" is a

trademark of its respective owner.

Frequently asked questions

What is proxmox-mcp-suite?

proxmox-mcp-suite is Proxmox MCP Suite

How do I install proxmox-mcp-suite?

Open the GitHub repository and follow its README. Most MCP servers are added to your client's MCP config, then called by your agent.

Is proxmox-mcp-suite open source?

Yes — it is hosted on GitHub at https://github.com/ahmetem/proxmox-mcp-suite and has 3 stars.

Related MCP tools

IvanMurzakUnity-MCP

AI Skills, MCP Tools, and CLI for Unity Engine. Full AI develop and test loop. Use cli for quick setup. Efficient token usage, advanced tools. Any C# method may be turned into a tool by a single line. Works with Claude Code, Gemini, Copilot, Cursor and any other absolutely for free.

4,137 C#
aiai-integrationgame-development+16
jgravellejcodemunch-mcp

Cut AI token costs 95%+ on code exploration. The leading MCP server for precise, symbol-level GitHub code retrieval via tree-sitter AST. Works with Claude Code, Cursor & any MCP client. 313B+ tokens saved.

2,651 Python
claudeclaude-codeai-coding+17
iFuryStopen-codex-computer-use

👾 Open Computer Use – Open-Source Alternative to Codex Computer Use

1,899 Swift
computer-usecodex-computer-useopen-computer-use+17
riponcmprojectmem

Open-source coding agent memory. Records issues, attempts, fixes and decisions, then warns your agent before it repeats an approach that already failed. Native MCP server for Claude Code, Cursor, Antigravity and Codex. 100% local, no cloud, no telemetry. MIT.

796 Python
ai-agentsai-memoryai-tools+17
AVIDS2memorix

Open-source cross-agent memory layer for coding agents via MCP. Compatible with Claude Code, Codex, Cursor, Windsurf, Gemini CLI, Antigravity, OpenClaw, Hermes Agent, Oh-my-Pi, Pi, Copilot, Kiro, OpenCode, and Trae.

721 TypeScript
ai-codingclaude-codecopilot+17
agentic-boxmemora

Give your AI agents persistent, collective memory — with deduplicating absorb, supersession lineage, semantic search, and a graph UI. Speaks MCP.

715 Python
ai-agentclaudeknowledge-graph+13

Run your own MCP server? See who uses it and what to fix.

Measure it with TrackMCP