Manage Proxmox VE clusters through AI assistants — VMs, containers, and more
The MCP server io.github.antonio-mello-ai/mcp-proxmox enables AI assistants to manage Proxmox VE clusters through the Model Context Protocol. It provides capabilities to provision, manage, and monitor Proxmox infrastructure via natural language, including VMs and containers, plus operations like snapshot management and storage browsing.
🛠️ Key Features
Manage Proxmox VE clusters via AI assistants
Create and manage VMs
Create and manage containers
Snapshot management
Storage browsing
🚀 Use Cases
Provision Proxmox infrastructure using natural language
Monitor Proxmox environments through an MCP-connected assistant
Perform common admin tasks such as creating VMs/containers and managing snapshots
⚡ Developer Benefits
Supports MCP integration for tooling and assistant workflows
Relevant for virtualization and homelab automation contexts
Topic-aligned for model-context-protocol and mcp-server discovery
⚠️ Limitations
Source excerpt does not specify exact tool count, APIs, or detailed operational constraints.
MCP server for managing Proxmox VE clusters through AI assistants like Claude, Cursor, and Cline.
Provision, manage, and monitor your entire Proxmox infrastructure through natural language. Create VMs and containers, manage snapshots, browse storage, and more.
Quick Start
bash
# Run directly with uvx (no install needed)
uvx mcp-proxmox
# Or install with pip
pip install mcp-proxmox
Configuration
Set these environment variables (or create a .env file):
bash
PROXMOX_HOST=192.168.1.100 # Your Proxmox VE host
PROXMOX_TOKEN_ID=user@pam!mcp # API token ID
PROXMOX_TOKEN_SECRET=xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx # API token secret
Add to Cursor Settings > MCP with the same configuration as above.
Available Tools
Discovery
Tool
Description
list_nodes
List all cluster nodes with CPU, memory, and uptime
get_node_status
Detailed node info: CPU model, memory, disk, versions
list_vms
List QEMU VMs (filter by node or status)
list_containers
List LXC containers (filter by node or status)
get_guest_status
Detailed VM/CT status by VMID (auto-detects type and node)
Lifecycle
Tool
Description
start_guest
Start a stopped VM or container
stop_guest
Force-stop (requires confirmation)
shutdown_guest
Graceful ACPI/init shutdown
reboot_guest
Reboot (requires confirmation)
Storage
Tool
Description
list_storages
List storage pools with capacity and usage (filter by node)
list_storage_content
Browse ISOs, templates, backups, and disk images
Provisioning
Tool
Description
create_vm
Create a QEMU VM with configurable CPU, memory, disk, ISO, and network
create_container
Create an LXC container from a template
clone_guest
Clone a VM or CT (full or linked clone, cross-node support)
delete_guest
Permanently delete a stopped VM or CT (requires confirmation)
Backup & Restore
Tool
Description
list_backups
List backup files (filter by node, storage, or VMID)
create_backup
Create a vzdump backup (snapshot/suspend/stop modes, zstd/lzo/gzip)
restore_backup
Restore a VM or CT from a backup file (requires confirmation)
Command Execution
Tool
Description
exec_command
Run a command inside a QEMU VM via guest agent
Note:exec_command requires qemu-guest-agent installed and running inside the VM. Not supported for LXC containers (Proxmox API limitation).
Console Interaction
Tool
Description
vm_screenshot
Capture a PNG screenshot of a QEMU VM's display (login prompt, boot messages, installer, GUI)
vm_send_key
Send a key or combination to a VM console (requires confirmation)
vm_send_text
Type ASCII text into a VM console (US layout; requires confirmation)
Note: Console tools do not require qemu-guest-agent and need only the VM.Console privilege. vm_screenshot opens the same VNC websocket tunnel the Proxmox web UI uses (RFB 3.8 with DES-based VNC auth) and is implemented in pure stdlib (custom WebSocket client + DES). Only QEMU VMs are supported. Text entry is layout-dependent: vm_send_text maps ASCII to a US keyboard layout (Shift-mapping for uppercase and symbols); non-ASCII input (e.g. Cyrillic) is not supported by the sendkey protocol.
Snapshots
Tool
Description
list_snapshots
List all snapshots for a VM/CT
create_snapshot
Create a new snapshot
rollback_snapshot
Rollback to a snapshot (requires confirmation)
delete_snapshot
Delete a snapshot (requires confirmation)
Network
Tool
Description
list_networks
List bridges, bonds, and physical interfaces on a node
Resize
Tool
Description
resize_guest
Resize CPU, memory, and/or disk of a VM or container (requires confirmation)
Monitoring
Tool
Description
get_guest_metrics
CPU, memory, network, disk I/O over time
list_tasks
Recent tasks on a node (backups, migrations, etc.)
Firewall
Tool
Description
list_firewall_rules
List firewall rules for a VM/CT, node, or the cluster
add_firewall_rule
Add a firewall rule (action, direction, protocol, port, source/dest)
delete_firewall_rule
Delete a firewall rule by position (requires confirmation)
Migration
Tool
Description
migrate_guest
Live or offline migrate a VM/CT to another node (requires confirmation)
Templates & Cloud-init
Tool
Description
list_templates
List all VM templates available for cloning
create_template
Convert a stopped VM into a template (requires confirmation, irreversible)
configure_cloud_init
Set user, password, SSH keys, IP config, and DNS on a VM
Safety
Destructive operations (stop_guest, reboot_guest, rollback_snapshot, delete_snapshot, delete_guest, resize_guest, restore_backup, delete_firewall_rule, migrate_guest, create_template) require explicit confirm=true. The first call returns a warning describing the impact; only a second call with confirmation executes the action.
Examples
Once connected, you can ask your AI assistant:
"List all my VMs and their status"
"How much memory is VM 100 using?"
"Shut down container 105"
"Create a snapshot of VM 200 called before-upgrade"
"Show me the CPU usage of VM 100 over the last day"
"What tasks ran on node pve recently?"
"Which VMs are stopped?"
"What storage pools do I have and how full are they?"
"Show me available ISO images"
"Create a new Ubuntu VM with 4 cores and 8GB RAM"
"Clone VM 100 as a test environment"
"Create a Debian container from template"
"Delete the old test VM 999"
"Back up VM 100 to the zfs-backup-storage"
"Show me all backups for VM 200"
"Restore the latest backup of container 101"
"Run 'df -h' on VM 100"
"Check if nginx is running on VM 200"
"Take a screenshot of VM 100 so I can see the console"
"The VM is stuck at a login prompt — type 'root' and press Enter"
"Show me the network bridges on node pve"
"Give VM 100 more CPU — bump it to 8 cores"
"Add 50GB of disk to container 101"
"Show me the firewall rules on VM 100"
"Allow TCP port 443 on container 101"
"Migrate VM 200 to node pve2"
"List all templates in the cluster"
"Convert VM 102 into a template"
"Set cloud-init on VM 100: user admin, IP dhcp, DNS 8.8.8.8"
Troubleshooting
Missing required environment variables
This error means one or more required connection settings are not available to the server process. Copy .env.example to .env, then set PROXMOX_HOST, PROXMOX_TOKEN_ID, and PROXMOX_TOKEN_SECRET. If you configure an MCP client directly, make sure the same variables are present in that client's env block.
403 Permission check failed
A 403 usually means the API token is valid but does not have enough Proxmox privileges for the requested operation. Revisit Creating a Proxmox API Token and either uncheck "Privilege Separation" for full access or assign the specific privileges listed there, such as VM.Audit, VM.PowerMgmt, VM.Snapshot, or storage permissions for datastore operations.
exec_command fails for guest-agent or LXC reasons
exec_command only works for QEMU VMs with qemu-guest-agent installed and running inside the guest. If the tool reports that the QEMU guest agent is not responding, start or install the agent in the VM and retry. LXC containers are not supported by this Proxmox API path, so use another container access method instead.
ModuleNotFoundError: No module named 'mcp.server.fastmcp'
The MCP Python SDK 2.0 removed the mcp.server.fastmcp module that mcp-proxmox 1.2.1 and earlier import. Fresh installs (uvx mcp-proxmox, uv tool install, pip install) resolved mcp 2.x and failed on startup — in Claude Desktop this only shows up as "Server disconnected". Upgrade to mcp-proxmox 1.2.2 or later, which pins mcp<2. If you must stay on an older mcp-proxmox, constrain the SDK yourself: uvx --with "mcp<2" mcp-proxmox.
Self-signed certificate or connection errors
Proxmox commonly runs on port 8006, so check that PROXMOX_HOST and PROXMOX_PORT point to the same endpoint you use for the Proxmox web UI. For homelab clusters with self-signed certificates, leave PROXMOX_VERIFY_SSL=false or set it explicitly. Use PROXMOX_VERIFY_SSL=true only when the Proxmox certificate chains to a CA trusted by your runtime.
Development
bash
git clone https://github.com/felhen-ai/mcp-proxmox.git
cd mcp-proxmox
python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"# Run tests
pytest
# Lint
ruff check src/ tests/
ruff format src/ tests/
# Type check
mypy src/
License
MIT
Repository and registry compatibility
The canonical repository is https://github.com/felhen-ai/mcp-proxmox. Existing GitHub clone and issue URLs redirect to this repository. Do not recreate a repository at the old address, which would remove those redirects.
The canonical MCP Registry identifier is io.github.felhen-ai/mcp-proxmox. The legacy identifier io.github.antonio-mello-ai/mcp-proxmox remains supported with a separate server.legacy.json manifest. Both identifiers install the same mcp-proxmox package from PyPI, and the package name and CLI command remain unchanged. Keep both ownership markers in this README so future releases can be registered under either identifier.
New Docker publications use ghcr.io/felhen-ai/mcp-proxmox. Existing images and tags at ghcr.io/antonio-mello-ai/mcp-proxmox are retained for compatibility; they are not deleted or silently replaced. The legacy Glama badge above remains a discovery link until that directory migrates its entry.