Skip to main content
Glama

VMware MCP

A Model Context Protocol server for VMware vSphere. It gives an AI assistant read access to your vCenter Server or standalone ESXi host — inventory, performance counters, events and alarms — and, when you explicitly allow it, the ability to power VMs on and off, take snapshots, clone, reconfigure and migrate them.

It talks to vSphere with pyVmomi, the official VMware Python SDK, over the same vSphere Web Services API the vSphere Client uses. No agent, no appliance, and nothing to install on your hosts.

Read-only by default. Out of the box the server refuses every operation that would change your environment. Enabling writes is a deliberate, separate step.

Contents

Related MCP server: vSphere MCP Server

Quick start

Requires Python 3.10 or newer and network access to port 443 on vCenter or ESXi.

git clone https://github.com/ISH2YU/VMware-MCP.git
cd VMware-MCP
pip install -e .

Set the connection details and check that they work before wiring anything up to an AI client:

export VMWARE_HOST=vcenter.example.com
export VMWARE_USERNAME='svc-mcp@vsphere.local'
export VMWARE_PASSWORD='...'

vmware-mcp --check

--check logs in, prints the vCenter version, the account it authenticated as and the number of inventory objects it can see, then exits:

{
  "endpoint": "vcenter.example.com:443",
  "permission_mode": "read-only",
  "verify_ssl": true,
  "authenticated_as": "svc-mcp@vsphere.local",
  "server": {
    "name": "VMware vCenter Server 8.0.3 build-24022515",
    "product": "VMware vCenter Server",
    "version": "8.0.3",
    "build": "24022515",
    "api_version": "8.0.3.0",
    "api_type": "VirtualCenter",
    "os_type": "linux-x64",
    "vendor": "VMware, Inc.",
    "instance_uuid": "aaaa-bbbb-cccc",
    "license_product": "VMware VirtualCenter Server"
  },
  "inventory_objects_indexed": 214
}

If the TLS handshake fails, either point VMWARE_CA_BUNDLE at your vCenter's CA certificate or, for a lab, set VMWARE_VERIFY_SSL=false.

Then run the server. It speaks stdio by default, which is what desktop MCP clients expect:

vmware-mcp

Configuring an MCP client

Cursor

Add to ~/.cursor/mcp.json (or .cursor/mcp.json in a project):

{
  "mcpServers": {
    "vmware": {
      "command": "vmware-mcp",
      "env": {
        "VMWARE_HOST": "vcenter.example.com",
        "VMWARE_USERNAME": "svc-mcp@vsphere.local",
        "VMWARE_PASSWORD": "...",
        "VMWARE_PERMISSION_MODE": "read-only"
      }
    }
  }
}

Claude Desktop

Add to claude_desktop_config.json (macOS: ~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "vmware": {
      "command": "vmware-mcp",
      "env": {
        "VMWARE_HOST": "vcenter.example.com",
        "VMWARE_USERNAME": "svc-mcp@vsphere.local",
        "VMWARE_PASSWORD": "..."
      }
    }
  }
}

Over HTTP

To run the server once and share it, use the streamable HTTP transport:

vmware-mcp --transport streamable-http --host 127.0.0.1 --port 8000

The endpoint has no authentication of its own, so bind it to localhost or put it behind a proxy that does.

Permission modes

VMWARE_PERMISSION_MODE decides how much damage the server can do. Modes are cumulative.

Mode

What it allows

read-only (default)

Inventory, monitoring, events, alarms and performance counters. Every mutating tool refuses with an explanation.

write

Adds power operations, snapshot creation, clone, reconfigure and migrate.

destructive

Adds deleting VMs and reverting or deleting snapshots.

The mode is enforced server-side, before any call reaches vCenter, and it is stated in the server instructions so the model knows what it can attempt. Deleting a VM additionally requires an explicit confirm=true argument and refuses while the VM is powered on.

Belt and braces: give the service account only the vCenter privileges it needs. The permission mode protects against a confused model, not against a compromised one — vCenter roles are the real boundary.

Tools

All 27 tools are prefixed vsphere_.

Inventory

Tool

Purpose

vsphere_about

Product, version, API type, authenticated user and the active permission mode.

vsphere_list_datacenters

Every datacenter.

vsphere_list_clusters

Clusters with capacity, DRS and HA configuration.

vsphere_list_hosts

ESXi hosts with hardware, version and live CPU/memory utilisation.

vsphere_get_host

One host in full, including its VMs, datastores and networks.

vsphere_list_resource_pools

Reservations, limits and current usage.

vsphere_search_inventory

Search by name across object types when you don't know what a name refers to.

Virtual machines

Tool

Purpose

vsphere_list_vms

VMs filtered by name, power state, datacenter, cluster, host, guest OS or IP.

vsphere_get_vm

One VM in full: hardware, disks, NICs, guest networking and filesystems, snapshots.

vsphere_get_vm_summary_by_host

VM counts and vCPU/memory overcommitment per host.

Storage and networking

Tool

Purpose

vsphere_list_datastores

Capacity, free space, provisioned space and over-provisioning.

vsphere_list_networks

Standard port groups, distributed port groups (with VLAN and switch) and opaque networks.

Monitoring

Tool

Purpose

vsphere_list_tasks

Recent tasks, optionally scoped to an object.

vsphere_get_task

Poll a single task by id.

vsphere_list_running_tasks

What vCenter is doing right now.

vsphere_list_events

The audit trail: logins, changes, HA actions, hardware problems.

vsphere_list_alarms

Every currently triggered alarm, red first.

vsphere_get_performance

CPU, memory, disk and network counters for a VM or host, summarised.

Power, snapshots and lifecycle

Tool

Mode

Purpose

vsphere_change_vm_power_state

write

Power on/off, suspend, reset, or ask the guest to shut down or reboot.

vsphere_list_snapshots

read

Snapshot tree plus a flat list with paths.

vsphere_create_snapshot

write

Take a snapshot, optionally with memory or quiesced.

vsphere_revert_to_snapshot

destructive

Revert, discarding later changes.

vsphere_delete_snapshot

destructive

Delete one snapshot, a subtree, or all of them.

vsphere_clone_vm

write

Clone a VM or deploy from a template.

vsphere_reconfigure_vm

write

Change vCPU count, cores per socket, memory or notes.

vsphere_migrate_vm

write

vMotion, storage vMotion, or both.

vsphere_delete_vm

destructive

Delete a VM and its disks. Needs confirm=true.

Long operations (clone, migrate, snapshot) accept wait=false and return a task_id you can poll with vsphere_get_task. When waiting, progress is streamed to the client as MCP progress notifications.

Resources and prompts

Two resources:

  • vsphere://inventory/summary — counts and totals for the whole environment, useful as ambient context.

  • vsphere://vm/{identifier} — full detail for one VM, by name, moid, UUID or path.

Two prompts:

  • troubleshoot_vm(vm) — a diagnostic walkthrough for a slow, stuck or unreachable VM: performance counters for the VM and its host, recent events and tasks, alarms, snapshot age and datastore pressure.

  • capacity_report(scope) — cluster and host utilisation, overcommitment ratios, datastores nearing full, and a prioritised list of recommendations.

Configuration reference

Every setting is an environment variable; the flags shown override them.

Variable

Default

Description

VMWARE_HOST

required

vCenter Server or ESXi hostname or IP. Also VSPHERE_HOST.

VMWARE_USERNAME

required

vSphere account. Also VMWARE_USER, VSPHERE_USER.

VMWARE_PASSWORD

required

Password. Also VSPHERE_PASSWORD.

VMWARE_PORT

443

API port.

VMWARE_VERIFY_SSL

true

Verify the TLS certificate.

VMWARE_INSECURE

false

Shorthand for disabling verification.

VMWARE_CA_BUNDLE

—

CA certificate bundle to trust instead of the system store.

VMWARE_PERMISSION_MODE

read-only

read-only, write or destructive.

VMWARE_CONNECT_TIMEOUT

30

HTTP connection timeout in seconds.

VMWARE_TASK_TIMEOUT

600

How long to wait for a vSphere task before handing back its id.

VMWARE_MAX_RESULTS

500

Hard cap on items returned by any listing.

VMWARE_DEFAULT_PAGE_SIZE

100

Page size when a tool call omits limit.

VMWARE_CACHE_TTL

60

Seconds to cache the inventory tree used for path resolution.

VMWARE_MAX_CONCURRENCY

8

Maximum simultaneous calls to vCenter.

VMWARE_LOG_LEVEL

INFO

Log level. Logs go to stderr.

VMWARE_TRANSPORT

stdio

stdio, streamable-http or sse.

Command line flags: --vsphere-host, --vsphere-port, --username, --insecure, --ca-bundle, --permission-mode, --transport, --host, --port, --log-level, --check. Run vmware-mcp --help for details. There is deliberately no --password flag; passwords on a command line end up in the process list and shell history.

Referring to objects

Every tool that takes a VM, host, datastore, network or cluster accepts any of:

  • the object name — web-01, matched case-insensitively

  • a glob, where a name filter is accepted — web-*, db-0?

  • the managed object id — vm-1024, host-42

  • a UUID, for VMs and hosts — BIOS or instance UUID

  • the inventory path — /Prod/vm/Tier1/web-01, or any suffix of it

When several objects match, the tool lists the candidates with their managed object ids rather than picking one. Listings are paginated and report truncated, so a model can tell the difference between "that's all of them" and "there are more".

Security notes

  • Create a dedicated service account in vSphere rather than reusing an administrator. Grant it a read-only role for the default mode; add only the specific privileges you need if you enable writes.

  • The password is never returned by any tool. vsphere_about reports the connection settings with the password omitted.

  • Certificate verification is on by default. Disabling it is logged as a warning at startup.

  • Anything an MCP client can call, a model can call. Permission modes and the confirm flag on deletion exist because prompt injection through, for example, a VM annotation is a real risk. Keep the server in read-only unless a task genuinely needs more.

  • Secrets in client config files are stored in plain text by most MCP clients. Prefer a secret manager or environment variables where your client supports them.

Development

python -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

pytest                     # 200+ tests, no vCenter required
ruff check src tests       # lint
ruff format src tests      # format
mypy                       # type check

The test suite runs against an in-memory vCenter double (tests/fake_vsphere.py) that implements the two seams the server actually depends on: pyVmomi's SOAP stub and the PropertyCollector. Everything above those seams — the real client, the real property specs, the real mappers and the real tools — runs unmodified, so tests exercise the same code paths that talk to a live vCenter.

How it is put together

src/vmware_mcp/
├── config.py          Environment parsing and permission modes
├── errors.py          Error types, all with client-safe messages
├── server.py          MCPServer assembly, resources and prompts
├── cli.py             Argument parsing, --check, transports
├── vsphere/
│   ├── session.py     Connection, TLS, reconnect on session expiry
│   ├── client.py      Async facade; runs pyVmomi on a bounded thread pool
│   ├── query.py       PropertyCollector batching and inventory paths
│   ├── mappers.py     Pure vSphere-to-JSON translation
│   ├── lookup.py      Name/moid/UUID/path resolution
│   ├── tasks.py       Task polling with progress reporting
│   ├── perf.py        Performance counter queries
│   └── monitoring.py  Events, tasks and alarms
└── tools/             One module per area of vSphere

Two decisions worth knowing about:

Everything reads through the PropertyCollector. Touching managed object attributes one at a time costs a round trip each, which is unusable against a vCenter with thousands of VMs. A listing here is a single RetrievePropertiesEx call regardless of how many objects come back.

pyVmomi is synchronous, MCP is not. Blocking calls run on a thread pool bounded by VMWARE_MAX_CONCURRENCY, and vSphere tasks are polled from the event loop rather than blocking a thread, so progress can be streamed and a cancelled request does not strand a worker.

License

MIT. See LICENSE.

Available Tools

27 tools
vsphere_aboutA
Read-onlyIdempotent

Identify the connected vSphere endpoint.

Returns the product name and version of vCenter Server or ESXi, the API version, the account the server is authenticated as, and the permission mode this MCP server is running in. Call this first to find out whether write operations are allowed at all.

Inputs: no named parameters. Returns: object

vsphere_change_vm_power_stateA
Destructive

Change the power state of a virtual machine.

Prefer the guest actions when VMware Tools is running: shutdown_guest and reboot_guest let the operating system shut down cleanly, whereas power_off and reset are the equivalent of pulling the plug and can lose unwritten data.

Guest actions complete inside the guest and return as soon as vSphere has passed the request to VMware Tools, so wait does not apply to them.

Requires permission mode write or higher.

Args: vm: VM name, managed object id, UUID or inventory path. action: One of power_on, power_off, suspend, reset, shutdown_guest, reboot_guest or standby_guest. wait: Wait for the vSphere task to finish before returning. timeout_seconds: Override the default task timeout.

Inputs: vm (required), action (required), wait, timeout_seconds. Returns: object

vsphere_clone_vmA

Clone a virtual machine or deploy one from a template.

Placement defaults to the source VM's host, datastore, resource pool and folder; override any of them individually. Cloning copies every disk, so it can take a long time for large VMs -- pass wait=false to get a task id back immediately and poll it with vsphere_get_task.

Requires permission mode write or higher.

Args: vm: Source VM or template (name, moid, UUID or inventory path). name: Name for the new virtual machine. host: Target ESXi host. datastore: Target datastore for the clone's files. resource_pool: Target resource pool. folder: Target VM folder. power_on: Power the clone on once the clone completes. as_template: Register the clone as a template instead of a VM. wait: Wait for the clone task to finish. timeout_seconds: Override the default task timeout.

Inputs: vm (required), name (required), host, wait, +6 more. Returns: object

vsphere_create_snapshotA

Take a snapshot of a virtual machine.

Snapshots are not backups: they grow with every write and degrade performance if left in place, so delete them once they are no longer needed.

Requires permission mode write or higher.

Args: vm: VM name, managed object id, UUID or inventory path. name: Name for the new snapshot. description: Optional free-text description. include_memory: Capture the memory state too, so a revert returns the VM to a running state. Slower and larger. quiesce: Ask VMware Tools to quiesce the guest filesystem for a crash-consistent snapshot. Requires VMware Tools. wait: Wait for the snapshot task to finish. timeout_seconds: Override the default task timeout.

Inputs: vm (required), name (required), wait, quiesce, +3 more. Returns: object

vsphere_delete_snapshotA
Destructive

Delete one snapshot, a snapshot subtree, or every snapshot of a VM.

Deleting a snapshot consolidates its data into the parent disk; the VM keeps all current data. Requires permission mode destructive.

Args: vm: VM name, managed object id, UUID or inventory path. snapshot: Snapshot name, path or moid. Ignored when delete_all is set. remove_children: Also delete snapshots taken from this one. delete_all: Delete the entire snapshot tree ("Delete All" in the vSphere client). wait: Wait for the delete task to finish. timeout_seconds: Override the default task timeout. Consolidation of large snapshots can take a long time.

Inputs: vm (required), wait, snapshot, delete_all, +2 more. Returns: object

vsphere_delete_vmA
Destructive

Permanently delete a virtual machine and all of its files from disk.

This cannot be undone. The VM must be powered off. Requires permission mode destructive and an explicit confirm=true.

Args: vm: VM name, managed object id, UUID or inventory path. confirm: Must be true; guards against accidental deletion. wait: Wait for the delete task to finish. timeout_seconds: Override the default task timeout.

Inputs: vm (required), confirm (required), wait, timeout_seconds. Returns: object

vsphere_get_hostA
Read-onlyIdempotent

Full detail for one ESXi host, including attached VMs and datastores.

Args: host: Host name, managed object id (host-42), hardware UUID or inventory path.

Inputs: host (required). Returns: object

vsphere_get_performanceA
Read-onlyIdempotent

Read CPU, memory, disk and network performance counters for a VM or host.

Each counter comes back summarised (latest, average, minimum, maximum) over the requested window. realtime gives 20-second samples for roughly the last hour; the rollup intervals cover longer periods.

Args: entity: VM or host name, managed object id, UUID or inventory path. entity_type: Whether entity is a vm or a host. interval: realtime, 5min, 30min, 2hours or 1day. samples: Number of samples to retrieve, most recent first. counters: Specific counters in group.name.rollup form, e.g. cpu.ready.summation. Defaults to a standard set.

Inputs: entity (required), samples, counters, interval, +1 more. Returns: object

vsphere_get_taskA
Read-onlyIdempotent

Check the state and progress of a single vSphere task.

Tools that started work with wait=false, or that timed out waiting, return a task_id you can poll here.

Args: task_id: The task's managed object id, e.g. task-4218.

Inputs: task_id (required). Returns: object

vsphere_get_vmA
Read-onlyIdempotent

Full detail for one virtual machine.

Includes virtual hardware (disks and network adapters), guest networking and filesystem usage reported by VMware Tools, resource allocations, and the snapshot tree.

Args: vm: VM name, managed object id (vm-1024), BIOS/instance UUID or inventory path such as /Prod/vm/Tier1/web-01.

Inputs: vm (required). Returns: object

vsphere_get_vm_summary_by_hostA
Read-onlyIdempotent

Aggregate VM counts and allocated resources per ESXi host.

A quick capacity/placement overview: how many VMs sit on each host, how many are powered on, and how much vCPU and memory has been handed out.

Inputs: no named parameters. Returns: object

vsphere_list_alarmsA
Read-onlyIdempotent

List every currently triggered alarm across the inventory.

This is the fastest way to answer "what is unhealthy right now".

Args: status: Filter by severity: red (alert) or yellow (warning). include_acknowledged: Include alarms an operator has already acknowledged. limit: Maximum number of alarms to return. offset: Number of matches to skip, for paging.

Inputs: limit, offset, status, include_acknowledged. Returns: object

vsphere_list_clustersA
Read-onlyIdempotent

List compute clusters with their capacity, DRS and HA configuration.

Args: name: Optional filter on cluster name. Plain text matches as a case-insensitive substring; * and ? enable glob matching. datacenter: Only return clusters in this datacenter. limit: Maximum number of clusters to return. offset: Number of matches to skip, for paging.

Inputs: name, limit, offset, datacenter. Returns: object

vsphere_list_datacentersA
Read-onlyIdempotent

List every datacenter in the vSphere inventory.

Inputs: no named parameters. Returns: object

vsphere_list_datastoresA
Read-onlyIdempotent

List datastores with capacity, free space and over-provisioning.

provisioned_gib includes space promised to thin disks that has not been written yet, so it can exceed capacity_gib; overprovisioned flags exactly that case.

Args: name: Datastore name filter (substring, or glob with */?). datacenter: Only datastores in this datacenter. datastore_type: Filter by storage type such as VMFS, NFS or vsan. min_used_percent: Only datastores at or above this utilisation, for spotting the ones about to fill up. limit: Maximum number of datastores to return. offset: Number of matches to skip, for paging.

Inputs: name, limit, offset, datacenter, +2 more. Returns: object

vsphere_list_eventsA
Read-onlyIdempotent

List recent vSphere events, newest first.

Events are the audit trail of the environment: logins, configuration changes, HA actions, alarm transitions and hardware problems.

Args: entity: Restrict to events about this object and its children. entity_type: What kind of object entity names. hours: How far back to look, in hours. categories: Filter by info, warning, error or user. limit: Maximum number of events to return.

Inputs: hours, limit, entity, categories, +1 more. Returns: object

vsphere_list_hostsA
Read-onlyIdempotent

List ESXi hosts with hardware, version and live CPU/memory utilisation.

Args: name: Optional host name filter (substring, or glob with */?). cluster: Only return hosts belonging to this cluster. datacenter: Only return hosts in this datacenter. connection_state: Filter by connected, disconnected or notResponding. limit: Maximum number of hosts to return. offset: Number of matches to skip, for paging.

Inputs: name, limit, offset, cluster, +2 more. Returns: object

vsphere_list_networksA
Read-onlyIdempotent

List networks: standard port groups, distributed port groups and opaque networks.

Distributed port groups additionally report their VLAN configuration and the distributed switch they belong to.

Args: name: Network name filter (substring, or glob with */?). datacenter: Only networks in this datacenter. kind: standard-portgroup, distributed-portgroup or opaque-network. limit: Maximum number of networks to return. offset: Number of matches to skip, for paging.

Inputs: kind, name, limit, offset, +1 more. Returns: object

vsphere_list_resource_poolsA
Read-onlyIdempotent

List resource pools with their reservations, limits and current usage.

Args: name: Optional resource pool name filter. limit: Maximum number of pools to return. offset: Number of matches to skip, for paging.

Inputs: name, limit, offset. Returns: object

vsphere_list_running_tasksA
Read-onlyIdempotent

List the tasks vCenter is currently running or has just finished.

Inputs: no named parameters. Returns: object

vsphere_list_snapshotsA
Read-onlyIdempotent

List the snapshots of a virtual machine as both a tree and a flat list.

Each snapshot carries the moid needed to revert to or delete it, and is_current marks the snapshot the VM is currently running from.

Args: vm: VM name, managed object id, UUID or inventory path.

Inputs: vm (required). Returns: object

vsphere_list_tasksA
Read-onlyIdempotent

List recent vSphere tasks, newest first.

Use this to see what changed in the environment and whether an operation someone else started succeeded.

Args: entity: Restrict to tasks on this object and its children. entity_type: What kind of object entity names. hours: How far back to look, in hours. states: Filter by task state: queued, running, success or error. limit: Maximum number of tasks to return.

Inputs: hours, limit, entity, states, +1 more. Returns: object

vsphere_list_vmsA
Read-onlyIdempotent

List virtual machines with power state, resources and guest details.

All filters are combined with AND. Results are capped; check the truncated field and page with offset when it is true.

Args: name: VM name filter. Plain text matches as a case-insensitive substring; * and ? enable glob matching. power_state: poweredOn, poweredOff or suspended. datacenter: Only VMs in this datacenter. cluster: Only VMs whose host belongs to this cluster. host: Only VMs registered to this ESXi host (name or moid). guest_os: Filter on the configured guest OS name, e.g. Ubuntu. ip_address: Filter on the primary guest IP address (substring match). include_templates: Include VM templates alongside real VMs. only_templates: Return only VM templates. limit: Maximum number of VMs to return. offset: Number of matches to skip, for paging.

Inputs: host, name, limit, offset, +7 more. Returns: object

vsphere_migrate_vmA

Move a virtual machine to another host, datastore or resource pool.

Supplying host performs a vMotion, datastore a storage vMotion, and both at once a combined migration. At least one target is required.

Requires permission mode write or higher.

Args: vm: VM name, managed object id, UUID or inventory path. host: Destination ESXi host. datastore: Destination datastore. resource_pool: Destination resource pool. priority: vMotion scheduling priority. wait: Wait for the migration task to finish. timeout_seconds: Override the default task timeout.

Inputs: vm (required), host, wait, priority, +3 more. Returns: object

vsphere_reconfigure_vmA

Change a virtual machine's CPU count, memory size or notes.

Unless CPU/memory hot-add is enabled on the guest, vSphere requires the VM to be powered off for resource changes and will reject the request otherwise. Only the supplied fields are changed.

Requires permission mode write or higher.

Args: vm: VM name, managed object id, UUID or inventory path. cpu_count: New total number of virtual CPUs. cores_per_socket: Cores per virtual socket. Must divide cpu_count. memory_mb: New memory size in MiB. annotation: Replacement notes/annotation text. wait: Wait for the reconfigure task to finish. timeout_seconds: Override the default task timeout.

Inputs: vm (required), wait, cpu_count, memory_mb, +3 more. Returns: object

vsphere_revert_to_snapshotA
Destructive

Revert a virtual machine to a snapshot, discarding later changes.

Everything written since the snapshot was taken is lost. Requires permission mode destructive.

Args: vm: VM name, managed object id, UUID or inventory path. snapshot: Snapshot name, snapshot path (parent/child) or snapshot moid. Defaults to the VM's current snapshot. suppress_power_on: Leave the VM powered off after reverting. wait: Wait for the revert task to finish. timeout_seconds: Override the default task timeout.

Inputs: vm (required), wait, snapshot, timeout_seconds, +1 more. Returns: object

vsphere_search_inventoryA
Read-onlyIdempotent

Search the inventory by name across several object types at once.

Useful when you know a name but not what kind of object it is, or when you need the managed object id to pass to another tool.

Args: query: Name to search for (substring, or glob with */?). types: Object types to search. Defaults to all of vm, host, cluster, datastore, network, datacenter, resource_pool and folder. limit: Maximum number of matches to return in total.

Inputs: query (required), limit, types. Returns: object

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4.1/5.0

    Scored across 27 tools

    Disambiguation4/5

    Most tools target distinct resources and actions, but vsphere_list_running_tasks overlaps with vsphere_list_tasks when filtered by state, creating some ambiguity. Overall, boundaries are clear and well-documented.

    Naming Consistency5/5

    All tools follow a consistent vsphere_<verb>_<noun> pattern using lowercase snake_case. Verbs like list, get, create, delete, and change are used predictably across the set.

    Tool Count2/5

    With 27 tools, the count exceeds the 25-tool threshold for 'too many' and feels heavy. While each tool addresses a specific vSphere concern, the inclusion of overlapping task-listing tools adds unnecessary bulk.

    Completeness4/5

    The tool set covers the core vSphere lifecycle: inventory queries, VM management (clone, reconfigure, migrate, delete, power), snapshots, tasks, events, alarms, and performance. Minor gaps like having no direct datastore/cluster creation are acceptable and don't block common workflows.

    Maintenance

    ActivityInactive
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers