一个 Model Context Protocol 服务器，用于向 Claude Code 和其他 MCP 客户端公开 VMware vCenter / ESXi 虚拟机生命周期管理工具。基于 pyVmomi 构建。
列出 vCenter 数据中心（按主机分组）或独立 ESXi 主机上的虚拟机
创建虚拟机（优先网络启动；支持精简或厚置备；ESXi 目标支持嵌套虚拟化选项）
虚拟机开关机
删除虚拟机（若正在运行则先关机，然后从磁盘销毁）
查找时可接受显示名称或 moref ID（例如 vm-42）——使用 moref 路径可跳过清单扫描，在大型环境中速度更快。
Related MCP server: VMWare MCP
Python 3.10 或更高版本
可通过网络访问的 vCenter Server 或独立 ESXi 主机
拥有执行所需操作权限的 vSphere 账户（list_vms 仅需只读权限；创建/删除操作需要相应的虚拟机和资源池权限）
安装到项目本地的虚拟环境中。使用 venv 可以将 vcenter-mcp 及其依赖项（主要是 pyVmomi）与系统 Python 隔离开来。
从本仓库克隆后安装：
python3 -m venv .venv
.venv/bin/pip install --upgrade pip
.venv/bin/pip install -e .开发环境安装（同时安装 pytest）：
.venv/bin/pip install -e ".[dev]"在本 README 中，命令均使用
.venv/bin/...。你也可以在每个 shell 中执行一次source .venv/bin/activate并省略前缀——效果相同。
使用虚拟环境的 Python 运行交互式设置：
.venv/bin/python -m vcenter_mcp setup系统将提示你输入：
目标名称（例如 lab-vcenter）——用于后续引用此目标
vCenter / ESXi 的主机名或 IP
用户名和密码
目标类型：vcenter 或 esxi
(仅 vCenter) 数据中心和集群名称
数据存储名称
一个或多个网络配置文件，每个包含一个名称以及一个或多个端口组名称
设置程序会将配置写入 ~/.config/vcenter-mcp/config.json（模式 0600）。你可以随时重新运行以添加新目标或更新现有目标。
{
"default_target": "lab-vcenter",
"targets": {
"lab-vcenter": {
"host": "vcenter.lab.example.com",
"user": "admin@vsphere.local",
"password": "...",
"type": "vcenter",
"datacenter": "Lab DC",
"cluster": "Lab Cluster",
"datastore": "datastore1",
"networks": {
"standard": ["VM Network"],
"secure-boot": ["pg-secure-1", "pg-secure-2"]
},
"default_network": "standard"
}
},
"templates": {
"esxi": { "cpu": 4, "ram_mb": 16384, "disk_gb": 100, "disk_provisioning": "thin", "guest_id": "vmkernel7Guest", "vhv": true },
"ubuntu": { "cpu": 2, "ram_mb": 4096, "disk_gb": 40, "disk_provisioning": "thin", "guest_id": "ubuntu64Guest", "vhv": false },
"rhel": { "cpu": 2, "ram_mb": 4096, "disk_gb": 40, "disk_provisioning": "thin", "guest_id": "rhel9_64Guest", "vhv": false }
}
}网络配置文件是端口组的列表；第一个条目将成为启动网卡。要添加自定义虚拟机类型，请在 templates 中添加条目——传递给 create_vm 的 vm_type 字符串将与此字典进行匹配。
使用虚拟环境的 Python 通过绝对路径注册 MCP 服务器。Claude Code 会在一个全新的 shell 中启动服务器，该 shell 不会 继承你已激活的 venv，因此必须使用绝对路径——在此处指向普通的 python 将导致无法导入 vcenter_mcp。
VCENTER_MCP_DIR="$(pwd)" # run this from the repo root, after install
claude mcp add --scope user vcenter -- "$VCENTER_MCP_DIR/.venv/bin/python" -m vcenter_mcp或者直接内联你想要的绝对路径：
claude mcp add --scope user vcenter -- /absolute/path/to/vcenter-mcp/.venv/bin/python -m vcenter_mcp读取工具（list_vms）可以安全地允许而不进行提示。将其添加到 ~/.claude/settings.json 中的 permissions.allow：
{
"permissions": {
"allow": [
"mcp__vcenter__list_vms"
]
}
}破坏性工具（create_vm、power_on_vm、power_off_vm、delete_vm）默认不在允许列表中——Claude 会在每次调用时提示你。
工具 | 功能 |
| 列出目标上的虚拟机。vCenter 目标按数据中心内的主机分组；ESXi 目标列出主机上的所有内容。 |
| 创建一个优先网络启动的虚拟机。选择一个 |
| 通过显示名称或 moref ID 开启虚拟机。 |
| 通过显示名称或 moref ID 强制关闭虚拟机。 |
| 永久删除虚拟机（若正在运行则先关机，然后从磁盘销毁）。 |
vcenter-mcp 使用未验证的 SSL 上下文进行连接，这与 govc 和大多数 pyVmomi 示例代码使用的默认设置相同，因为实验室 vCenter 通常使用自签名证书。如果你的目标使用正确签名的证书，并且你希望进行真正的验证，请替换 src/vcenter_mcp/client.py 中的 _ssl_context()。
python3 -m venv .venv
.venv/bin/pip install -e ".[dev]"
.venv/bin/pytest测试在 CI 中的 Python 3.10、3.11 和 3.12 上运行（参见 .github/workflows/test.yml）。
Create a VM that network boots first. vm_type: esxi, ubuntu, rhel (or any type defined in config templates). disk_provisioning: thin (default) or thick. network_profile: named profile from target config (e.g. standard, secure-boot).
| Name | Required | Description | Default |
|---|---|---|---|
| name | Yes | ||
| vm_type | Yes | ||
| target | No | ||
| network_profile | No | ||
| cpu | No | ||
| ram_mb | No | ||
| disk_gb | No | ||
| disk_provisioning | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description carries the burden but only mentions 'network boots first' as a behavioral trait. It does not disclose whether the operation is idempotent, requires authentication, or what happens on conflict.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is very concise, using four short lines with no unnecessary words. Each sentence adds value and is front-loaded with the primary purpose.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
Given 8 parameters and no output schema details, the description is incomplete: it omits required fields, resource constraints, and 5 parameters entirely. An output schema exists but does not compensate for missing parameter explanations.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, and the description only explains three parameters (vm_type, disk_provisioning, network_profile) out of eight. Critical parameters like name, cpu, ram_mb, and disk_gb are left undocumented.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the verb 'Create' and the resource 'VM', and adds specificity with 'network boots first'. Sibling tools are management operations, so there is no confusion with alternatives.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description gives examples for vm_type and disk_provisioning defaults, but does not provide explicit guidance on when to use this tool versus alternatives, nor does it mention prerequisites or constraints.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Permanently delete a VM (power off if running, then destroy from disk). Accepts display name or moref ID (e.g. 'vm-42').
| Name | Required | Description | Default |
|---|---|---|---|
| name_or_id | Yes | ||
| target | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description shoulders full transparency. It discloses the irreversible destructive action, the pre-step (power off), and the input format. Missing details like permission requirements or error behavior, but core behavioral traits are present.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Two sentences with no extraneous information. Action and input format are front-loaded, making it efficient and clear.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description covers the action and key parameter but lacks explanation for the optional 'target' parameter and does not mention output schema or return behavior, leaving gaps for a complete understanding.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 0%, so the description must explain all parameters. It clarifies that name_or_id accepts a display name or moref ID, but completely omits the 'target' parameter, leaving users without guidance on its purpose or usage.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description explicitly states 'Permanently delete a VM' and details the process ('power off if running, then destroy from disk'), clearly differentiating it from sibling tools like list_vms or power_on_vm.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description explains how to identify the VM (name or moref ID) but does not provide explicit guidance on when to use this tool versus alternatives, such as powering off or creating a snapshot.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
List VMs on a target.
Standalone ESXi: lists all VMs on the host.
vCenter: groups VMs by host within the specified datacenter (defaults to the target's configured datacenter).
| Name | Required | Description | Default |
|---|---|---|---|
| target | No | ||
| datacenter | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description discloses significant behavioral differences between standalone ESXi and vCenter, including optional datacenter grouping. Without annotations, it carries the full burden and does so well. It implies read-only operation (list) but does not explicitly state lack of side effects.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is concise with two bullet points, front-loading the main action. Every sentence provides value, no redundancy.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
The description covers the tool's core behavior and parameter roles. With an output schema present (context signal), return values need not be described. It is fairly complete for a list tool, though explicit mention of read-only nature would strengthen it further.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
With 0% schema description coverage, the description must compensate. It explains the 'datacenter' parameter defaults to the target's configured datacenter but lacks detail on the 'target' parameter (e.g., format, valid values). Some meaning added, but insufficient given the gap.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the tool lists VMs on a target, with specific behavior for standalone ESXi vs vCenter. It uniquely identifies the resource (VMs) and action (list), distinguishing it from sibling tools like create_vm, delete_vm, etc.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description provides clear context for when to use the tool (listing VMs on a target) and mentions key parameters (target, datacenter). However, it does not explicitly state when not to use or suggest alternative tools, though siblings are distinct actions.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Hard power off a VM by display name or moref ID (e.g. 'vm-42').
| Name | Required | Description | Default |
|---|---|---|---|
| name_or_id | Yes | ||
| target | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
With no annotations, the description should disclose more behavioral details. It states 'hard power off' implying forceful shutdown but does not mention risks (data loss), prerequisites, or behavior if VM is already off or not found.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single concise sentence with no wasted words, but lacks structure such as separate sections for usage or parameters.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple tool with an output schema, the description is adequate for the core action, but fails to explain the optional parameter and lacks behavioral context, leaving gaps for an AI agent.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, so the description must explain parameters. It explains name_or_id (display name or moref ID) but completely ignores the 'target' parameter, leaving its purpose unclear.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the action (hard power off) and the resource (VM), and specifies the identification methods (display name or moref ID), distinguishing it from siblings like power_on_vm or delete_vm.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description implies when to use (to hard power off a VM) but provides no explicit guidance on when not to use, such as preferring a soft shutdown or prerequisites like the VM being powered on.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Power on a VM by display name or moref ID (e.g. 'vm-42').
| Name | Required | Description | Default |
|---|---|---|---|
| name_or_id | Yes | ||
| target | No |
| Name | Required | Description |
|---|---|---|
| result | Yes |
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
The description only states the action without disclosing side effects (e.g., idempotency), required permissions, or error conditions. With no annotations, the agent lacks critical behavioral context for a mutation tool.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
A single, front-loaded sentence with no wasted words. Every piece of information earns its place.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a simple tool with two parameters and an output schema, the description covers the primary parameter but misses behavioral details (e.g., what happens if VM is already on) and the purpose of the 'target' parameter. Adequate but with clear gaps.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
The description explains that name_or_id accepts a display name or moref ID, adding value beyond the schema structure. However, the optional 'target' parameter is not explained, and schema coverage is 0%, so the description only partially compensates.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description clearly states the action ('Power on a VM') and the resource identifiers ('by display name or moref ID'), which distinguishes it from siblings like power_off_vm.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The description specifies how to identify the VM (name or moref ID) with an example, but does not provide guidance on when to choose one identifier over the other or mention when not to use the tool.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
v0.1.0create_vmdelete_vmlist_vmspower_off_vmpower_on_vmScored across 5 tools
Each tool has a clearly distinct purpose: create, delete, list, power on, power off. No overlap in functionality.
All tools follow a consistent verb_noun pattern in snake_case (create_vm, delete_vm, list_vms, power_off_vm, power_on_vm). Minor plural variation for list_vms is acceptable.
5 tools cover the essential VM lifecycle operations without being excessive or insufficient for a vCenter MCP server.
Missing common operations like get single VM details, update VM configuration, clone, or snapshot management. Basic CRUD and power actions are present but gaps exist for full lifecycle management.
Provides capabilities that let LLM agents perform a range of infrastructure management tasks.
Interact with the Stitch API using natural language commands.
Manage Rackspace Spot Kubernetes Cloudspaces, node pools, and VMs from your AI assistant.
Interact with your Google Cloud Firestore resources using natural language commands.
