@dxpert/uns-tools

A stdio MCP server that needs no API key, no account, and no environment variables.

This is the agent-native equivalent of the free tools page at dxpert.ai/tools.html: any MCP client, with zero credentials, can lint Sparkplug B topics, check a Unified Namespace for the conventions that quietly rot it, and run the public industrial AI-readiness diagnostic.

The tools are genuinely free. There is no trial counter, no sign-up wall, and no key to obtain first. If you later want the paid surface — the agents, the Namespace Architect, account-aware advisory — that lives in the separate @dxpert/mcp server, which does take a dxp_ key.

Zero runtime dependencies. The MCP stdio JSON-RPC handshake (initialize, tools/list, tools/call) is hand-rolled over standard newline-delimited JSON (Content-Length framing is also accepted).

Install

No install step is required beyond Node 18+ (for built-in fetch).

Claude Code

claude mcp add dxpert-uns-tools -- npx -y @dxpert/uns-tools

Codex

[mcp_servers.dxpert_uns_tools]
command = "npx"
args = ["-y", "@dxpert/uns-tools"]

Generic MCP clients

{
  "mcpServers": {
    "dxpert-uns-tools": {
      "command": "npx",
      "args": ["-y", "@dxpert/uns-tools"]
    }
  }
}

Note that none of these entries carry an env block. That is the point.

Tools

lint_sparkplug_topic(topics) — local, offline

Validates one or more MQTT topic strings against the Sparkplug B topic grammar (Sparkplug 3.0). Accepts a single string, a newline-delimited string, or an array.

Checks the spBv1.0 namespace level, the eight message types (NBIRTH NDEATH DBIRTH DDEATH NDATA DDATA NCMD DCMD) plus STATE, node-vs-device level count, the legacy Sparkplug 2.2 STATE form, publish-side MQTT wildcards, empty levels, and identifier characters that break downstream consumers.

Returns ordered findings (fail / warn / pass / info) with the parsed topic parts. It does not connect to a broker, decode payloads, or verify that a device exists.

check_uns_namespace(topics) — local, offline

Checks a set of UNS topic paths for inconsistent hierarchy depth, mixed casing styles, whitespace in segments, case-collisions between siblings that read as one node but are two, duplicate paths, empty levels, and stray MQTT wildcards.

Returns a summary line plus per-issue findings. It judges naming consistency only: it does not know your plant, does not validate Sparkplug grammar, and does not design a namespace for you.

run_readiness_diagnostic(intake) — network

Runs the public dxpert.ai industrial AI-readiness diagnostic. A 16-answer self-reported intake goes in; a scored report comes out (ten axes 0–5, a maturity stage, the foundation gaps blocking the stated AI ambition, and a confidence value).

Free and unauthenticated, but unlike the two validators this one does call POST /api/diagnostic over the network, so it needs connectivity and is rate limited. Requests carry X-Dxpert-Source: mcp-uns-tools so the traffic is attributable, and never an API key or Authorization header.

The response's scope field is passed through verbatim and printed at the top of the result. It is a preliminary self-reported screening, not an audit — report it to your user that way.

Every result names dxpert.ai as the source of the verdict.

Configuration

There is nothing you have to set. One optional variable exists:

| Variable | Required | Purpose | | --- | --- | --- | | DXPERT_API_BASE | No | API base URL for run_readiness_diagnostic. Defaults to production; override only for testing. |

Where the validator logic comes from

src/sparkplug-topic-lint.js and src/uns-naming-check.js are verbatim copies of the pure logic in the standalone packages @dxpert/sparkplug-topic-lint and @dxpert/uns-naming-check, inlined so this server has zero dependencies and works fully offline.

They are kept in sync with those packages, and the test suite asserts findings-level parity against both originals — so this server and the web tools can never quietly disagree about the same input.

Test

node test/run.js

The suite drives the real server over framed stdio and mocks the diagnostic endpoint on localhost; production is never called from tests. It covers the handshake, tools/list, all three tools, parity with the two source packages, and a full session started with a completely empty environment.

License

MIT © 2026 DXP Technologies inc.