MCP server: One MCP server.

Nutanix Prism Central

@wyre-aiMCP server · External

Multitenant Streamable HTTP bridge over Nutanix's official v4 API MCP server (Prism Central).

MCP gateways & tool hubsv1.0.7Apache-2.0TypeScript
Freeno charge per call

Runs on your machine from its package; the steps are below.

What it does

  1. Reads
  2. Writes
  3. High impact

At most: only reads or looks up information.

What it costs

Free

Odel charges nothing per call. The publisher may charge for its own service.

Runs locally over stdio

This server isn't hosted — your MCP client launches it from a package registry. Use one of the commands below, or drop the config into your client (e.g. Claude Desktop).

Dockerghcr.io/wyre-ai/nutanix-mcp:v1.0.7

Run

docker run -i --rm ghcr.io/wyre-ai/nutanix-mcp:v1.0.7

MCP client config

{
  "mcpServers": {
    "nutanix-prism-central": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/wyre-ai/nutanix-mcp:v1.0.7"
      ]
    }
  }
}
README8 min read

nutanix-mcp

Multitenant Streamable HTTP bridge over nutanix/ntnx-api-mcp-server — Nutanix's official Prism Central v4 API MCP server — built so the WYRE conduit gateway can forward per-tenant Nutanix credentials as HTTP headers.

Upstream pin: Nutanix's server is a technical preview, pinned here at tag v0.8 (Apache-2.0). See Bumping the upstream pin.

Why

The upstream server is stdio-only (nutanix-mcp serve-stdio is its only serve mode) and reads its Prism Central credentials from environment variables at process startup — single-tenant per process. Our gateway is multi-tenant: every request carries the calling org's credentials as HTTP headers, and the vendor container has to translate those headers into something the upstream understands.

Because the upstream has no HTTP mode to proxy to, this bridge holds an MCP client session over stdio to each tenant child and re-serves it over Streamable HTTP:

  1. Listens on :8080 with POST /mcp and GET /health. Startup requires a real CONDUIT_S2S_SECRET. Every /mcp request must carry a valid X-Gateway-S2S signature of that secret.
  2. 401-gates every /mcp request on the X-Nutanix-Pc-* credential headers (below). Those headers are the only Prism Central credential source.
  3. Lazily spawns one nutanix-mcp serve-stdio child per credential tuple (keyed by a hash), with the tenant's PC_* env vars set, and connects an MCP client to it over stdio.
  4. Serves both protocol eras on /mcp via the v2 SDK's createMcpHandler(factory, { legacy: 'stateless' }) — 2025-era initialize-handshake clients (the conduit gateway today) and modern 2026-07-28 envelope clients. tools/list and tools/call delegate to the tenant's child session.
  5. Evicts idle children after 60 minutes (IDLE_EVICT_MS).

Tool names pass through unchanged: the upstream's 24 tools — 20 {namespace}_execute tools (aiops, clustermgmt, datapolicies, dataprotection, files, iam, licensing, lifecycle, microseg, monitoring, multidomain, networking, objects, opsmgmt, prism, security, storage, tenancy, vmm, volumes) plus 4 discovery tools (listOperations, getOperationSchema, getCodeSample, getOperationPermissions).

Read-only in v1 — deliberate

Every child is spawned with READ_ONLY_MODE=true (also the upstream default): the upstream rejects all non-GET operations before they ever reach Prism Central. v1 of this bridge ships read-only as a deliberate fleet decision. Write support would be a reviewed, versioned change to credentialsToChildEnv() in src/credentials.ts — not a config flip.

Credential contract

The gateway forwards these headers on every /mcp request; the bridge maps them onto the upstream child's environment. conduit's vendor-config must match this table exactly.

HeaderChild env varRequiredNotes
X-Nutanix-Pc-HostPC_HOSTyesPrism Central IP or FQDN
X-Nutanix-Pc-PortPC_PORTnoUpstream default 9440. Upstream quirk: any port other than 9440 makes it use http://, not https://
X-Nutanix-Pc-UsernamePC_USERNAMEwith passwordBasic auth pair
X-Nutanix-Pc-PasswordPC_PASSWORDwith usernameBasic auth pair
X-Nutanix-Pc-Api-KeyPC_API_KEYalternativeSent to PC as the X-ntnx-api-key request header; upstream prefers it over basic auth when both are set
X-Nutanix-Pc-InsecurePC_INSECUREno"true"/"false" — skip TLS verification (default false)

Validity rule: pcHost present AND (apiKey present OR username+password present). Anything else → HTTP 401 with a JSON-RPC error body.

READ_ONLY_MODE=true is additionally forced on every child (see above).

API-spec artifacts (baked at build time)

The upstream builds its tool surface from YAML API-spec artifacts, not from the live PC. nutanix-mcp init downloads them — and without PC credentials it runs in latest_release mode against the public developers.nutanix.com namespace API (no PC access required; verified empirically: 20 namespaces). The Docker build runs init once and bakes the artifacts into the image at /opt/nutanix-mcp/artifacts, shared read-only by all tenant children. Consequences:

  • tools/list and the discovery tools work with no reachable PC — only {namespace}_execute calls touch Prism Central.
  • Artifact versions are the latest public release at image build time, not the tenant PC's exact versions (the upstream's pc_compatible mode would need live PC access at spawn). For the read-only v1 surface this is the right trade: shared artifacts, fast tenant spawns.

Configuration

Copy .env.example. Leave CONDUIT_S2S_SECRET empty in that file and set a real value only in the environment you run.

Env varDefaultNotes
CONDUIT_S2S_SECRETempty (refuses to start)Service-to-service secret. The HTTP server logs an error and exits non-zero when this is empty or a known placeholder (changeme, secret, replace-me, your-s2s-secret, replace-with-a-real-secret, and the other example values). Generate one with openssl rand -hex 32. Compose will not start without it. The value is never logged.
MCP_ALLOW_INSECURE_DEVunsetSet to exactly 1 only for local development to start HTTP without CONDUIT_S2S_SECRET. The process logs a warning, skips X-Gateway-S2S, and binds 127.0.0.1 only. A non-loopback MCP_HTTP_HOST refuses to start. A known placeholder secret still refuses to start.
MCP_HTTP_HOST0.0.0.0 when a real secret is setBind address. A configured secret honors this value, including when it is unset (historical 0.0.0.0 container bind). The image and Compose set 0.0.0.0 inside the container. With no secret, MCP_ALLOW_INSECURE_DEV=1 binds 127.0.0.1 only.
PORT8080Public listen port (MCP_HTTP_PORT is the fallback).
NUTANIX_MCP_DIR/opt/nutanix-mcpUpstream checkout (venv + artifacts).
NUTANIX_MCP_BIN$NUTANIX_MCP_DIR/.venv/bin/nutanix-mcpUpstream console script the bridge spawns.
ARTIFACTS_DIR$NUTANIX_MCP_DIR/artifactsBaked YAML API-spec artifacts.
CHILD_LOG_DIR/tmp/nutanix-mcp-logsWritable dir for upstream per-process log files.
IDLE_EVICT_MS3600000Idle tenant timeout (60 min).
SPAWN_TIMEOUT_MS60000Max wait for a child to answer the MCP handshake.

/health and /healthz stay unauthenticated and do not read credentials.

HTTP authentication

With CONDUIT_S2S_SECRET set, every /mcp request must include X-Gateway-S2S. The check is verifyS2sHeader in src/s2s-verify.ts: the header value is t=<unix seconds>,v1=<hex>, where the hex is HMAC-SHA256 of the literal string t=<unix seconds> keyed with CONDUIT_S2S_SECRET (64 lowercase hex characters). The timestamp must be within 300 seconds of the server clock. A missing or invalid header is HTTP 401, before Prism Central credentials are read.

Sign with the same CONDUIT_S2S_SECRET the server process is using. Local development exports that value in the shell; reuse it for this request instead of generating another one.

bash
TS=$(date +%s)
SIG=$(printf 't=%s' "$TS" | openssl dgst -sha256 -hmac "$CONDUIT_S2S_SECRET" -hex | awk '{print $NF}')
curl -sS http://127.0.0.1:8080/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -H "X-Gateway-S2S: t=${TS},v1=${SIG}" \
  -H 'X-Nutanix-Pc-Host: pc.example.com' \
  -H 'X-Nutanix-Pc-Api-Key: fake' \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-03-26","capabilities":{},"clientInfo":{"name":"dev","version":"0"}}}'

This bridge is gateway mode. /mcp also requires X-Nutanix-Pc-Host and either X-Nutanix-Pc-Api-Key or both X-Nutanix-Pc-Username and X-Nutanix-Pc-Password. Those headers are the only credential source for that request. PC_* values in the process environment are not used as a fallback.

Local development

bash
# 1. Get the upstream at the pinned tag with a venv + artifacts
git clone --branch v0.8 --depth 1 https://github.com/nutanix/ntnx-api-mcp-server ../ntnx-api-mcp-server
cd ../ntnx-api-mcp-server
uv venv .venv && uv pip install .
ARTIFACTS_DIR=$PWD/artifacts .venv/bin/nutanix-mcp init   # no PC creds needed
cd -

# 2. Build and run the bridge against it
npm ci && npm run build && npm test
# HTTP refuses to start until CONDUIT_S2S_SECRET is a real value.
# Export it so this shell keeps the same secret for the server, the
# signing request above, and the Docker commands below.
export CONDUIT_S2S_SECRET="$(openssl rand -hex 32)"
NUTANIX_MCP_DIR=../ntnx-api-mcp-server node dist/index.js

# Alternative: local HTTP only, without a service-to-service secret (always
# binds 127.0.0.1). A non-loopback MCP_HTTP_HOST makes this exit. The empty
# assignment applies only to this process; the exported secret stays in the shell.
MCP_ALLOW_INSECURE_DEV=1 CONDUIT_S2S_SECRET= \
  NUTANIX_MCP_DIR=../ntnx-api-mcp-server node dist/index.js

# 3. Smoke it (health is unauthenticated; /mcp needs X-Gateway-S2S — see above)
curl -s localhost:8080/health

Or build the image (it bakes everything, including a stdio smoke test that fails the build if serve-stdio can't answer tools/list). The image listens on 0.0.0.0 inside the container and still refuses to start without a real CONDUIT_S2S_SECRET. Publish the port on loopback. Prism Central credentials arrive on each request as X-Nutanix-Pc-* headers.

bash
docker build --platform linux/amd64 -t ghcr.io/wyre-ai/nutanix-mcp:dev .
# Pass the exported secret. An empty value makes the container exit.
docker run --rm \
  -e CONDUIT_S2S_SECRET="$CONDUIT_S2S_SECRET" \
  -p 127.0.0.1:8080:8080 \
  ghcr.io/wyre-ai/nutanix-mcp:dev

Compose refuses to start when CONDUIT_S2S_SECRET is unset. The gateway service publishes 127.0.0.1:8080:8080 and does not inject PC_HOST, PC_USERNAME, PC_PASSWORD, PC_API_KEY, or env_file. Use the exported secret so a later signing request matches the container:

bash
docker compose up

Bumping the upstream pin

The upstream is pinned to the reviewed tag v0.8 in the Dockerfile (NUTANIX_MCP_REF) — never main (NSA MCP guidance / fleet security baseline). To bump:

  1. Review the upstream diff between the current pin and the new tag (tool surface, credential handling, READ_ONLY_MODE semantics).
  2. Change NUTANIX_MCP_REF in the Dockerfile and the tag in this README.
  3. docker build locally — the build-time smoke tests assert the venv entrypoint runs and the stdio tool surface still answers (update the expected tool count if namespaces changed).
  4. Land as a feat:/fix: PR so semantic-release cuts a version.

License

Apache-2.0. The bundled ntnx-api-mcp-server is Apache-2.0 by Nutanix.

Similar modules

Browse MCP gateways & tool hubs