VMware Pilot

VMware Pilot

Author: Wei Zhou, VMware by Broadcom — wei-wz.zhou@broadcom.com This is a community-driven project by a VMware engineer, not an official VMware product. For official VMware developer tools see developer.broadcom.com.

English | 中文

Multi-step workflow orchestration for VMware MCP skills — state machine, approval gates, audit trail.

Companion skills handle everything else:

Skill	Scope	Install
vmware-aiops	VM lifecycle, deployment, guest ops, cluster	uv tool install vmware-aiops
vmware-monitor	Read-only: inventory, health, alarms, events	uv tool install vmware-monitor
vmware-storage	Datastores, iSCSI, vSAN management	uv tool install vmware-storage
vmware-vks	Tanzu Namespaces, TKC cluster lifecycle	uv tool install vmware-vks
vmware-nsx	NSX networking: segments, gateways, NAT	uv tool install vmware-nsx-mgmt
vmware-nsx-security	DFW firewall rules, security groups	uv tool install vmware-nsx-security
vmware-aria	Aria Ops: metrics, alerts, capacity	uv tool install vmware-aria
vmware-avi	AVI load balancing, pool management, AKO K8s ops	uv tool install vmware-avi

Install

uv tool install vmware-pilot
vmware-pilot mcp          # start the MCP server (stdio)

Offline / Air-Gapped Install (from source)

This project uses the modern PEP 517 build system (hatchling), so there is no setup.py by design — that is expected, not a missing file. If you cloned the source and hit ERROR: File "setup.py" or "setup.cfg" not found ... editable mode currently requires a setuptools-based build, your pip is older than 21.3 and cannot do an editable (-e) install with a non-setuptools backend. Editable mode is a developer convenience, not needed to run the tool — do one of:

# From the source tree — a normal (non-editable) install builds a wheel:
pip install .              # NOT  pip install -e .

# ...or upgrade pip first, and editable works too:
pip install --upgrade pip && pip install -e .

For a truly air-gapped host, build the wheels on a connected machine and copy them over — the target then needs no network:

# On a connected machine, collect this package + its dependencies as wheels:
pip wheel . -w dist        # → dist/*.whl   (or: uv build, for just this package)

# Copy dist/ to the air-gapped host, then install offline:
pip install --no-index --find-links dist vmware-pilot

MCP Tools (13 — 4 read, 9 write)

Built-in Templates (15)

n is the number of VMs (or drift items); ranges depend on which optional parameters are set. clone_and_test needs snapshot_name (the snapshot to clone from), resizes staging while it is powered off, and has 8 steps when change_spec is a guest command (an extra gate before it runs in staging). investigate_alert's approvals are synthesis checkpoints; its steps are all reads. See skills/vmware-pilot/references/templates.md for parameters and steps.

Upgrading to 2.0

2.0 renames one MCP tool: plan_workflow → create_workflow_from_template. It always created and saved a workflow from a template (it is a write, not a dry run); the new name says so and sets it apart from create_workflow, which saves your own step list. Parameters and results are unchanged, and so are the CLI commands. The skill catalog also follows vmware-vks's renames (get_supervisor_kubeconfig → issue_supervisor_kubeconfig, get_tkc_kubeconfig → issue_tkc_kubeconfig). Workflows saved with the old names are translated when loaded. If an agent prompt, script or ~/.vmware/rules.yaml names plan_workflow, update it. Requires vmware-policy ≥ 1.19.0. Details: RELEASE_NOTES.md.

MCP Configuration

{
  "mcpServers": {
    "vmware-pilot": {
      "command": "vmware-pilot",
      "args": ["mcp"]
    }
  }
}

Fallback: {"command": "uvx", "args": ["--from", "vmware-pilot", "vmware-pilot-mcp"]} still works, but uvx re-resolves against PyPI on every start and fails behind a TLS-inspecting corporate proxy (invalid peer certificate: UnknownIssuer). The installed entry point above touches the network zero times; set UV_NATIVE_TLS=true if you must use uvx.

Pilot runs the steps itself: --dispatch

By default pilot executes nothing: run_workflow returns each step in pending_dispatch and your agent performs it. Add --dispatch (or set VMWARE_PILOT_DISPATCH=stdio) and pilot runs each step itself, in that skill's own MCP server (vmware-<skill>-mcp, which must be installed), so the skill's policy rules, confirmation gate and audit apply as usual. Nothing after an approval gate runs before approve, and workflows reach completed.

{"mcpServers": {"vmware-pilot": {"command": "vmware-pilot", "args": ["mcp", "--dispatch"]}}}

stdio only: pilot refuses to start with --dispatch and --transport https/http (the skill servers it starts would have no caller identity, so roles could not apply). A step that takes longer than VMWARE_PILOT_DISPATCH_TIMEOUT seconds (default 1800) is recorded interrupted — its effect is unknown — and rollback waits for you to check it. run_workflow and approve take as long as the steps they run: set your MCP client's tool timeout above your longest step. If a call does time out, Pilot carries on and answers nothing else until the step ends; then get_workflow_status shows the outcome. Never perform those steps yourself.

Remote access (HTTPS) and limiting tools

By default vmware-pilot mcp speaks stdio to one local client and offers every tool. Two options change that (vmware-policy ≥ 1.21.0).

Serve over HTTPS so clients on other machines can connect. Every caller needs a bearer token:

vmware-audit token create alice          # prints the token once; give it to alice
vmware-audit token list                  # names only
vmware-audit token revoke alice          # refused from its next request, no restart
vmware-pilot mcp --transport https --host 0.0.0.0 --port 8443 \
    --cert /etc/vmware-mcp/server.pem --key /etc/vmware-mcp/server-key.pem \
    --public-name mcp.example.com

Connect a client (Claude Code shown; any MCP client that supports streamable HTTP and custom headers works):

claude mcp add --transport http vmware-pilot https://mcp.example.com:8443/mcp \
    --header "Authorization: Bearer $VMWARE_MCP_TOKEN"

{
  "mcpServers": {
    "vmware-pilot": {
      "type": "http",
      "url": "https://mcp.example.com:8443/mcp",
      "headers": { "Authorization": "Bearer ${VMWARE_MCP_TOKEN}" }
    }
  }
}

• The server will not start without a token, refuses Host/Origin values outside --public-name, and checks before it binds that unauthenticated requests get 401. Use a certificate your clients trust for those names.
• Behind a TLS-terminating reverse proxy on the same machine use --transport http --host 127.0.0.1 --public-name mcp.example.com and pass the Authorization header through.
• Behind a load balancer or gateway on another machine, keep --transport https (a self-signed certificate is fine), list in --public-name the host name the balancer sends as Host, and have it forward Authorization. Its health probe can use GET /healthz, which needs no token and returns only {"status": "ok"} (vmware-policy ≥ 1.23.0).
• Send traceparent or X-Request-ID with each request to have it stored on the audit row; vmware-audit log --correlation-id ID then finds the calls behind one platform request.
• The audit log records each caller as token:<name>. Calls naming different targets run in parallel; calls that name no target run one at a time, so name the target when several people share a server.
• Everyone who can reach the endpoint acts with this server's account. Tokens identify callers; they grant nothing on the platform. Run one endpoint per privilege level.
• Several people or roles on one endpoint: run vmware-audit roles enable once, then give each token a role (vmware-audit token create NAME --role reader, or a role in roles.yaml naming tools and targets; vmware-policy ≥ 1.24.0). A role limits what the token may call on this endpoint; it cannot grant more than the account has. See vmware-policy's README, "Roles for network callers".

Offer only read tools, plus the writes you choose:

vmware-pilot mcp --expose read --allow-write cancel_workflow
# or: VMWARE_PILOT_MCP_EXPOSE=read VMWARE_PILOT_MCP_ALLOW_WRITE=cancel_workflow vmware-pilot mcp

Tools not offered do not appear in the client's tool list. A bad value or an unknown write tool stops the server (exit 2). This limits what clients are offered; it is not authorization, so the account itself should be least-privilege.

Details: skills/vmware-pilot/references/setup-guide.md ("Serving MCP over HTTPS", "Limiting which tools are offered").

License

MIT