dxpert.ai — free UNS tools (no API key)

ai.dxpert由 dxpert.ai 發布v0.1.4更新於 Nov 1, 2026

Free Sparkplug B topic linting, UNS naming checks and AI-readiness screening. No API key.

已驗證STDIO僅桌面開發者工具資料與分析

概覽

AI 產生的概覽

免費且免金鑰的 MCP 伺服器,可檢查 Sparkplug B MQTT 主題、檢視統一命名空間(UNS)命名一致性,並執行工業 AI 就緒度診斷。

可以做什麼
  • 依 Sparkplug B(3.0)文法驗證 MQTT 主題字串,包含訊息類型、節點/裝置層級與萬用字元
  • 檢查一組 UNS 主題路徑的層級深度不一致、大小寫混用、空白字元、大小寫衝突、重複路徑與空層級
  • 送出 16 題自評問卷,取得包含十個軸向評分、成熟度階段、基礎落差與信心值的就緒度報告
適用情境
適合在助理處理工業物聯網或 MQTT 設定、broker 封包或程式碼時,核對 Sparkplug 主題或維持統一命名空間的命名規範;也適合需要快速取得自評式 AI 就緒度初篩的情況。
執行需求
透過 npx 以 stdio 在本機執行,需要 Node 18 以上版本;不需帳號、API 金鑰或環境變數。就緒度診斷會經網路呼叫 dxpert.ai 的 API,並受速率限制;可選變數 DXPERT_API_BASE 可覆寫 API 基礎位址,僅供測試使用。
安裝前請注意
兩個檢查工具在本機執行,不會把資料送到任何地方;但就緒度診斷會把問卷答案經網路送到 dxpert.ai,請求帶有 X-Dxpert-Source 標頭。其結果是自評式初步篩檢而非稽核,應據實說明。檢查工具不會連線 broker,也不會驗證裝置是否存在。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 dxpert.ai — free UNS tools (no API key),將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

@dxpert/uns-tools

[M8ven Score]

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. When you want an agent to read your own data, that lives in the separate @dxpert/mcp server, which takes a dxp_ key. A free key includes dxpert Advisor (a monthly question allowance) and Try Pro: 5 runs on any agent included in dxpert Pro. dxpert Pro includes every UNS agent, dxpert Advisor without the monthly count, and the Namespace Architect when it is released. Current plans: dxpert.ai/store.

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

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

Codex

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

Generic MCP clients

json
{  "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:

VariableRequiredPurpose
DXPERT_API_BASENoAPI 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

sh
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.

來源:README.md,提交 11aa7ba

工具

3

lint_sparkplug_topic

Lint MQTT topics against the Sparkplug B grammar

Validate one or more MQTT topic strings against the Sparkplug B topic grammar (Sparkplug 3.0). Free, no API key, no account: the grammar runs locally inside this server and nothing is sent anywhere. Call this whenever a topic that is meant to be Sparkplug appears in a config, a broker trace, a code review, or a user question -- it is faster and far more reliable than reasoning about the grammar from memory. It 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. It returns ordered findings (fail / warn / pass / info) with the parsed topic parts, not a yes/no. It does NOT connect to a broker, decode payloads, or check whether the device actually exists.

輸入參數topics必填

check_uns_namespace

Check UNS topic paths for convention problems

Check a set of Unified Namespace (UNS) topic paths for the convention problems that quietly rot a namespace: inconsistent hierarchy depth, mixed casing styles across segments, whitespace in segments, case-collisions between siblings that read as one node but are two, duplicate paths, empty levels, and stray MQTT wildcards. Free, no API key, no account: the checks run locally inside this server and nothing is sent anywhere. Call this when reviewing a proposed namespace, an ISA-95-style hierarchy, a broker topic dump, or a tag export -- before anyone builds on top of it, because renaming a namespace later is the expensive part. It returns a summary line plus per-issue findings (fail / warn / pass / info). It judges naming consistency only: it does NOT know your plant, does not validate Sparkplug grammar (use lint_sparkplug_topic for that), and does not design a namespace for you.

輸入參數topics必填

run_readiness_diagnostic

Run the public AI-readiness diagnostic

Run the public dxpert.ai industrial AI-readiness diagnostic: a 16-answer self-reported intake in, a scored report out (ten axes 0-5, a maturity stage, the foundation gaps blocking the stated AI ambition, and a confidence value). Free and requires no API key or account, but unlike the two validators this one DOES call the dxpert.ai API over the network, so it needs connectivity and is rate limited. Call this when someone asks how ready a plant or site is for AI/analytics, what is blocking them, or where to start -- and you can supply honest answers to the intake fields. Ask the user for the values you do not have rather than guessing them; the verdict is only as good as the intake. The response carries a "scope" field which this tool passes through verbatim: it is a preliminary self-reported screening, not an audit, and should be reported to the user as such.

輸入參數intake必填

版本歷史

3
  1. v0.1.4最新Nov 1, 2026
  2. v0.1.2Oct 4, 2026
  3. v0.1.0Sep 16, 2026

評分

已滿足適用檢查項的 90%

該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通25 分
  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具必需15 分
  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分