vcf-ops-mcp

An MCP server that wraps the VCF Operations (vROps) suite-api REST API, exposing resources, metrics, and alerts as MCP tools so an LLM client can query monitored infrastructure directly.

Setup

python3 -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"
cp .env.example .env   # then fill in your VCF Operations details

Required configuration (via .env or real environment variables):

The server acquires a token from /suite-api/api/auth/token/acquire on first use, caches it, and transparently re-acquires it when it's near expiry or rejected with 401.

Server transport/auth configuration (also via .env or real environment variables):

Running

By default this runs as a standalone remote server over streamable-http, bound to 127.0.0.1:8000, requiring a bearer token on every request:

export VCFOPS_MCP_BEARER_TOKEN="$(openssl rand -hex 32)"
vcf-ops-mcp
# or
python -m vcf_ops_mcp

GET /healthz is unauthenticated (for load balancer/orchestrator liveness checks); everything else requires the bearer token. 127.0.0.1 only listens locally — to actually reach it from another host, bind VCFOPS_MCP_HOST=0.0.0.0 (or run it behind a reverse proxy) and make sure the bearer token is the only thing standing between the network and credentials capable of querying your whole monitored environment, so treat it like any other secret and prefer TLS termination (e.g. a reverse proxy) in front of it rather than plaintext HTTP over an untrusted network.

When VCFOPS_MCP_HOST isn't 127.0.0.1/localhost, FastMCP's own DNS-rebinding protection (a check against the incoming request's Host header) has nothing to allowlist by default, since it only auto-configures that allowlist for a loopback host. Left unset, no Host-header check is enforced and the bearer token is your only gate — fine on a network you trust, but set VCFOPS_MCP_ALLOWED_HOSTS to the hostname(s)/IP:port clients actually connect through (comma-separated) for defense in depth on a shared or untrusted network.

Point an MCP client at it as a streamable-http server, e.g. in Claude Code:

claude mcp add --transport http vcf-operations http://<host>:8000/mcp \
  --header "Authorization: Bearer <your-token>"

Running over stdio instead

For local use where an MCP client spawns the server itself as a subprocess (no network exposure needed), set VCFOPS_MCP_TRANSPORT=stdio — the bearer token is not required in this mode. Example Claude Desktop config:

{
  "mcpServers": {
    "vcf-operations": {
      "command": "/absolute/path/to/.venv/bin/vcf-ops-mcp",
      "env": {
        "VCFOPS_MCP_TRANSPORT": "stdio",
        "VCFOPS_BASE_URL": "https://ops.example.com",
        "VCFOPS_USERNAME": "admin",
        "VCFOPS_PASSWORD": "changeme"
      }
    }
  }
}

Tools

Resources

• list_resources — search monitored objects by name/adapter kind/resource kind/health

• get_resource — full detail for one resource

• get_resource_properties — collected configuration properties

• get_resource_relationships — parent/child objects

Metrics

• list_metric_keys — available statKeys for an adapter/resource kind pair

• get_latest_stats — most recent metric value(s) for a resource

• query_stats — historical time series across resources, with rollup/interval

Alerts

• list_alerts — active/all alerts, optionally scoped to a resource or criticality

• get_alert — full detail for one alert

• list_alert_definitions — the rule definitions behind alerts

Testing

pip install -e ".[dev]"
pytest

Tests mock the VCF Operations HTTP API with respx — no live instance required.

Notes

• Pinned to mcp<2.0.0: the MCP Python SDK's 2.x line renamed FastMCP to MCPServer and moved it to mcp.server.mcpserver. This project targets the well-established 1.x mcp.server.fastmcp.FastMCP API.