libvirt-mcp
Enables management of KVM/QEMU virtual machines on remote libvirt hosts via SSH, with tools for inspection, lifecycle management, snapshots, and cloning.
README
libvirt-mcp
A Model Context Protocol (MCP) server for managing KVM/QEMU virtual machines on remote
libvirt hosts. Talks libvirt RPC over SSH — no agent on the hypervisor, no need to
install anything on the target host beyond the standard libvirtd.
Functionally a "virt-manager for LLMs": inspect, start/stop, snapshot, clone, and delete KVM/QEMU domains through 13 MCP tools.
Tools
| Group | Tools |
|---|---|
| Inspection | list_domains, domain_info, snapshot_list, screenshot |
| Lifecycle | start_domain, shutdown_domain (ACPI + poll), reboot_domain, force_destroy_domain |
| Snapshots | snapshot_create, snapshot_revert, snapshot_delete |
| Management | clone_from_template (linked clone), delete_domain (with optional wipe_disks) |
Destructive operations (force_destroy_domain, snapshot_revert, snapshot_delete,
delete_domain) require an explicit confirm=True argument so an LLM can't fire them
by accident.
Architecture
┌──────────────┐ ┌──────────────────────┐
│ MCP client │ ── stdio (JSON-RPC) ──▶ │ libvirt-mcp │
│ Claude Code, │ │ (Docker container) │
│ Desktop, … │ │ • libvirt-python │
└──────────────┘ │ • mcp SDK │
└──────────┬───────────┘
│
│ qemu+ssh://user@host/system
│ (uses host's ~/.ssh/)
▼
┌──────────────────────┐
│ Hypervisor │
│ • libvirtd │
│ • qemu-kvm │
└──────────────────────┘
The MCP server runs inside a Docker container (so libvirt-dev headers don't need to
be installed on your machine). It bind-mounts your ~/.ssh read-only and uses standard
libvirt SSH transport, inheriting ~/.ssh/config, agent forwarding, jump hosts, etc.
Setup
1. Pull the image
The container image is published as ghcr.io/rzippert/libvirt-mcp (public — no auth
required, even though the source repo is private):
docker pull ghcr.io/rzippert/libvirt-mcp:latest
Tags published:
latest— head ofmainX.Y.ZandX.Y— from git tagsvX.Y.Z(thevprefix is dropped per Docker convention)main,sha-<short>— for traceability
2. Create your config
mkdir -p ~/.config/libvirt-mcp
cat > ~/.config/libvirt-mcp/config.toml <<'EOF'
default_profile = "primary"
[profiles.primary]
uri = "qemu+ssh://user@hypervisor.example.com/system"
description = "Primary KVM hypervisor"
EOF
$EDITOR ~/.config/libvirt-mcp/config.toml
The libvirt URI is standard. SSH keys come from your host's ~/.ssh/. Make sure you
can already connect:
ssh user@hypervisor.example.com virsh -c qemu:///system list --all
3. Wire it up to an MCP client
Claude Code
claude mcp add libvirt --scope user -- \
docker run -i --rm \
-v "$HOME/.ssh:/home/ubuntu/.ssh:ro" \
-v "$HOME/.config/libvirt-mcp/config.toml:/etc/libvirt-mcp/config.toml:ro" \
--network host \
-e LIBVIRT_MCP_CONFIG=/etc/libvirt-mcp/config.toml \
ghcr.io/rzippert/libvirt-mcp:latest
Verify with claude mcp list. Remove with claude mcp remove libvirt --scope user.
Claude Desktop
Edit claude_desktop_config.json:
- macOS:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"libvirt": {
"command": "docker",
"args": [
"run", "-i", "--rm",
"-v", "/Users/youruser/.ssh:/home/ubuntu/.ssh:ro",
"-v", "/Users/youruser/.config/libvirt-mcp/config.toml:/etc/libvirt-mcp/config.toml:ro",
"--network", "host",
"-e", "LIBVIRT_MCP_CONFIG=/etc/libvirt-mcp/config.toml",
"ghcr.io/rzippert/libvirt-mcp:latest"
]
}
}
}
Replace /Users/youruser with your home directory. On Windows use forward slashes or
double-escape backslashes. Restart Claude Desktop after editing.
VS Code (GitHub Copilot Chat MCP)
Create .vscode/mcp.json in your workspace (project-scoped) or add to user settings:
{
"servers": {
"libvirt": {
"type": "stdio",
"command": "docker",
"args": [
"run", "-i", "--rm",
"-v", "${userHome}/.ssh:/home/ubuntu/.ssh:ro",
"-v", "${userHome}/.config/libvirt-mcp/config.toml:/etc/libvirt-mcp/config.toml:ro",
"--network", "host",
"-e", "LIBVIRT_MCP_CONFIG=/etc/libvirt-mcp/config.toml",
"ghcr.io/rzippert/libvirt-mcp:latest"
]
}
}
}
VS Code expands ${userHome} and ${workspaceFolder}. Open the Copilot Chat panel →
MCP servers → enable libvirt.
Configuration reference
~/.config/libvirt-mcp/config.toml (or override with LIBVIRT_MCP_CONFIG):
default_profile = "primary" # used when a tool is called without `profile=`
[profiles.<name>]
uri = "qemu+ssh://user@host/system" # standard libvirt URI
description = "Free-form description for humans"
Config search order:
$LIBVIRT_MCP_CONFIGenv var~/.config/libvirt-mcp/config.toml./config.dev.toml./config.toml
Safety model
- Allow-list validation on domain names (
^[a-z0-9][a-z0-9-]{0,30}$), bridge names, and absolute paths (no..segments). - Inspection tools open read-only libvirt connections; state-changing tools open read-write connections only when invoked.
- Confirmation gates on destructive operations:
force_destroy_domain,snapshot_revert,snapshot_delete,delete_domainall requireconfirm=True. - No subprocess on the hypervisor: every operation is pure libvirt RPC (no
qemu-img, novirt-install, novirshshell-out). Disk volumes are created and deleted through libvirt's storage volume API, reusing any active dir-pool that covers the target path or transparently creating a transient one.
Development
git clone git@github.com:rzippert/libvirt-mcp.git
cd libvirt-mcp
cp config.example.toml config.dev.toml && $EDITOR config.dev.toml
docker compose build
docker compose run --rm dev bash
Inside the dev container, source on the host (bind-mounted at /workspace) shadows the
installed package via PYTHONPATH, so edits take effect immediately. Exercise tools
directly:
python -c "from libvirt_mcp.tools.readonly import list_domains; \
import json; print(json.dumps(list_domains(), indent=2))"
Project layout:
libvirt_mcp/
├── __main__.py # python -m libvirt_mcp → mcp.run() (stdio)
├── server.py # FastMCP instance
├── connections.py # config loader + connect() context manager
├── states.py # VIR_DOMAIN_* → string
└── tools/
├── readonly.py # inspection
├── lifecycle.py # start/shutdown/reboot/force_destroy
├── snapshot.py # snapshot CRUD (revert/delete gated)
└── management.py # clone_from_template, delete_domain
Publishing happens automatically via .github/workflows/publish.yml on push to main
and on v* tags (multi-arch: linux/amd64, linux/arm64).
License
GNU Affero General Public License v3.0 or later (AGPL-3.0-or-later). See LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。