原型 · 预览版:收录官方注册中心里的全部服务器,每 30 分钟同步一次。使用 Google 登录即可撰写评价。原型 · 每 30 分钟更新
PopNx MCP为使用 AI 的人准备的 MCP 服务器 · 用大白话讲清楚 登录

AI Governance Engineer

✓ 服务方 · aigovernanceengineer.com

作者:losanchos5版本 0.6.1i★ 4 (星标属于其他仓库)i在云端运行官方网站 ↗
注册名称com.aigovernanceengineer/reference

安装方法

能做什么

Read-only AI governance reference: obligations, crosswalks, controls, glossary, patterns, chapters.

作者提供的原文

  • 连接方式 在云端运行
  • 账号或密钥 无需密钥
  • 分类 其他

为什么排在这里

此项在“连接 AI 的应用”标签中,该标签不使用星标,按以下顺序排列

1. 在 Claude 连接器目录中否
2. 团队实测过否
3. 适合泰国用户否
4. 不计星标的使用量20/100热度 0 · 新鲜度 12 · 仍在响应 8 · 用户喜欢 +0
5. 登记日期2026-11-18

用户喜欢: 0 · 最多加 5 分(每人 1 分),不能用钱买。

质量: 基础检查通过 4/5 项(仅展示,不用于排序)

安装

选择你使用的 AI 应用,然后复制设置。

用一句话让 AI 帮你安装

请帮我安装 AI Governance Engineer。按照 https://mcp.popnx.com/i/aigovernanceengineer-reference.md 操作,先把命令给我看,等我同意后再运行。

粘贴到能在你电脑上工作的 AI(如 Claude Code、Cursor)。它会读取我们的步骤,先给你看命令,等你同意。无需密钥,无需注册。 · 查看 AI 会读取的全文

通过 4/5 项检查 · 你的提问会经过开发者的服务器(mcp.aigovernanceengineer.com) · ✓ 命令已检查 2026年11月18日

第一步:你用哪个 AI?

选择你用的应用,步骤会显示在下方。

可连接 Claude · 约 2 分钟

不需要密钥,不用安装程序 · 在 claude.ai 或桌面应用设置一次,即可在聊天中使用 · 免费账户可添加 1 个 · Team/Enterprise 方案由所有者(Owner)添加

  1. 打开 Claude 的 Connectors 页面

    CustomizeConnectors

    CustomizeConnectors1Add2https://…3Add4
  2. 点击“Add” → “Custom” → “Web”,填写这 2 项

    把这里的内容复制到 Claude 中同名的输入框,然后点击 Continue

    Name · 名称AI Governance Engineer
    Remote MCP server URL · 地址https://mcp.aigovernanceengineer.com/mcp

    在 Claude 检测到的登录设置页点击 Continue · Authentication 按作者说明选择(Sign in now、Sign in when needed 或 No sign in),然后点击 Add

  3. 在聊天中开启

    +ConnectorsAI Governance Engineer

  4. 试一试

    用 AI Governance Engineer 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Claude 说正在使用 AI Governance Engineer,或回答中出现 AI Governance Engineer 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同

能否连接 ChatGPT 取决于方案

完整 MCP:网页版 ChatGPT Business、Enterprise 或 Edu · Pro:以读取/获取权限连接,修改数据的工具可能用不了 · Plus:官方文档尚未确认,如果在插件页看到添加自定义 MCP 的选项可以试试 · 仅限网页

  1. 在网页版 ChatGPT 打开添加 MCP 的页面

    Plugins+Add custom MCP server

    ChatGPTPlugins1+2https://…3Create4
  2. 填写名称和地址

    输入名称,在 Connection 下粘贴地址,勾选风险提示,然后点击 Create as a plugin

    Name · 名称AI Governance Engineer
    Connection · 地址https://mcp.aigovernanceengineer.com/mcp
  3. 在聊天中选择 AI Governance Engineer

    开启新聊天,输入前先从插件列表中选择 AI Governance Engineer

  4. 试一试

    用 AI Governance Engineer 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 ChatGPT 说正在使用 AI Governance Engineer,或回答中出现 AI Governance Engineer 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 OpenAI 帮助页面核实 · 你的界面可能因版本略有不同

可连接 Claude Code · 约 2 分钟

不需要密钥,不用安装程序

  1. 打开终端

    Mac:按 Command+空格,输入 Terminal · Windows:打开 PowerShell

  2. 粘贴这条命令并按 Enter

    claude mcp add --transport http --scope user reference https://mcp.aigovernanceengineer.com/mcp
  3. 在 Claude Code 中输入 /mcp

    应能看到 AI Governance Engineer 在列表中并显示已连接

  4. 试一试

    用 AI Governance Engineer 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Claude Code 说正在使用 AI Governance Engineer,或回答中出现 AI Governance Engineer 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同

可连接 Cursor · 约 2 分钟

不需要密钥,不用安装程序

  1. 点击这个按钮打开 Cursor

    在 Cursor 中打开

    Cursor 会请你确认安装,点击 Install

  2. 在 Cursor 中以 Agent 模式打开 AI 聊天

  3. 试一试

    用 AI Governance Engineer 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Cursor 说正在使用 AI Governance Engineer,或回答中出现 AI Governance Engineer 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Cursor 帮助页面核实 · 你的界面可能因版本略有不同

可连接 VS Code · 约 2 分钟

不需要密钥,不用安装程序

  1. 点击这个按钮打开 VS Code

    在 VS Code 中打开

    VS Code 会请你确认安装,点击 Install

  2. 在 VS Code 中以 Agent 模式打开 AI 聊天

  3. 试一试

    用 AI Governance Engineer 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 VS Code 说正在使用 AI Governance Engineer,或回答中出现 AI Governance Engineer 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 VS Code 帮助页面核实 · 你的界面可能因版本略有不同

其他支持 MCP 的应用(如 Windsurf、Zed、Gemini CLI)添加方式各不相同,请在应用列表中查看

查看所有支持 MCP 的应用 →

通过网页连接的应用使用的地址

地址https://mcp.aigovernanceengineer.com/mcp

JSON 设置在下方“开发者专用”中

卡住了?问问正在用 AI Governance Engineer 的人💬 讨论 →
开发者专用:终端命令和 JSON 设置

通过网页连接

Claude Code

claude mcp add --transport http --scope user reference https://mcp.aigovernanceengineer.com/mcp

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "reference": {
      "url": "https://mcp.aigovernanceengineer.com/mcp"
    }
  }
}

VS Code (mcp.json)

{
  "servers": {
    "reference": {
      "type": "http",
      "url": "https://mcp.aigovernanceengineer.com/mcp"
    }
  }
}

这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同,请以 README 为准。

一览

GitHub 星标i
4 (星标属于其他仓库)
此条目指向他人的仓库,因此星标不计入排名
每周下载量i
不在 npm/PyPI 上
已验证域名
✓ aigovernanceengineer.com
许可证i
CC-BY-4.0
连接方式
在云端运行
传输方式
streamable-http
最新版本i
0.6.1
分类i
其他
发布到注册中心i
2026年11月18日
代码最后更新i
2026年11月18日
最近检查
2026年11月18日
作者
losanchos5
注册状态
活跃

哪些应用可以使用它

依据这个服务器的连接方式和各应用自己的文档,检查于 2026年11月10日。

通过互联网连接,适用于

ClaudeChatGPTGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatOpen WebUIRaycast

全部应用及其优缺点 →

它能做什么

工具名称和描述与服务器在 2026年11月18日 报告的完全一致。这些内容由作者撰写,不是我们写的;我们以纯文本形式显示。

  • search_glossary

    Search the canonical terms of AI governance engineering (definitions from the Body of Knowledge glossary, chapter 09). Matches term names, acronyms and definition text; every query word must match. Returns term slugs for get_term, definitions, the canonical page URL of each term

  • get_term

    Get one glossary term by slug (e.g. "abstention-band"), by its id ("t-abstention-band"), by its page URL or by its name ("Abstention band", or an acronym in brackets such as "AUP"). Returns the definition, the canonical page, the chapters that use it and the other terms whose def

  • get_obligations

    Filter the obligation register (obligation → artefact → stack layer, with stable ids such as AIGE-OBL-EUAIA-ART9). All filters are optional and combine with AND. framework: an id, short name or name ("eu-ai-act", "EU AI Act", "ISO 42001", "nist"); role: a duty holder such as "pro

  • get_obligation

    Get one row of the obligation register by its stable id (e.g. "AIGE-OBL-EUAIA-ART9", any case) or its page URL: the requirement, the artefact that evidences it, the stack layers, duty holder, application date and later milestones, the patterns that implement it, the crosswalk ref

  • map_clause

    Map one clause of a framework to the others through the topic crosswalk (25 governance topics against the EU AI Act, GPAI Code, GDPR, ISO/IEC 42001 and related standards, NIST AI RMF, CSA AICM, OWASP, Korea, UK, Singapore, Council of Europe, OECD, G7, CEN-CENELEC and the Chinese

  • list_patterns

    List the reusable governance patterns of the Body of Knowledge (chapter 05), optionally only those of one stack layer (1 Govern-as-Code, 2 Inventory & Transparency, 3 Evals & Red Teaming as Evidence, 4 Runtime Controls & Observability, 5 Assurance & Continuous Compliance; a patte

  • get_pattern

    Get one governance pattern by slug ("policy-card", "kill-switch-circuit-breaker"), chapter 05 id ("pattern-aibom"), page URL or title: its layer, summary, the frameworks it maps to, the obligation ids that list it and the full text of its page (objectives, context, problem, solut

  • list_templates

    List the templates-and-schemas library: JSON Schemas (with a filled example and a human Markdown template) for the records a governed AI lifecycle produces, from the use-case record to the decommissioning runbook, and the organisation-wide policy kit (AI policy in YAML, prose and

  • get_template

    Get one entry of the templates-and-schemas library by name ("policy-card", "impact-assessment", "incident-record", "ai-policy", "raci"), file name or URL: what it records, its lifecycle stage and stack layers, the clauses it helps evidence, and the content of its files. part pick

  • search_bok

    Search the 24 chapters of the AI Governance Engineer Body of Knowledge and its pattern pages: chapter titles, summaries and key points, section headings and section text. Every query word must match. Returns chapters, sections (with the URL of the heading anchor) and patterns, be

  • list_controls

    List the reference controls of the open control profiles (draft control specifications with stable ids such as AIGE-CTL-EVAL-002, open for technical review). All filters are optional and combine with AND: profile (a profile slug such as "evaluation-environment", or its title), la

  • get_control

    Get one reference control of the open control profiles by its stable id (e.g. "AIGE-CTL-EVAL-002", any case, or "EVAL-002"), its page or JSON URL, or its title: objective, failure modes, scope, enforcement points, verification steps, the evidence it leaves (with schema ids), the

基础检查i

我们检查的是任何人都能看到的信息;我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月20日

  • ! 注意

    作者与仓库一致

    发布者(aigovernanceengineer.com)不是仓库所有者(losanchos5);可能是其他人代为上架的

  • ✓ 通过

    仍在维护

    代码最后更新于前天

  • ✓ 通过

    有许可证

    使用 CC-BY-4.0

  • ✓ 通过

    端点有响应

    在 151 毫秒内响应(HTTP 200)

  • – 不适用

    使用 OAuth 登录

    无需登录

  • ✓ 通过

    工具描述中没有隐藏指令

    已读取 12 个工具描述,未发现针对 AI 的隐藏指令

我们检查什么 → · JSON

作者:申请重新检查或报告信息错误
给作者:在 README 中加徽章

如果你是作者,可把这段代码贴到 README。徽章会链接回本页。

[![Featured on PopNx MCP](https://mcp.popnx.com/badge/aigovernanceengineer-reference.svg)](https://mcp.popnx.com/en/servers/aigovernanceengineer-reference)

讨论

提问、分享技巧或报告无法使用。按“有帮助”数量排序。

还没有帖子,欢迎第一个提问或分享。

此处内容由用户撰写,不代表 PopNx 团队。

登录后发帖

用户评价

还没有评价

登录后写评价

与“其他”中的其他服务器比较i

  • 星标数多于此分类中其他 20868 个服务器的 0%
  • 发布到注册中心的时间比此分类中其他 20868 个服务器的 97% 更新
  • 基础检查结果等于或优于此分类中其他 20868 个服务器的 60%

与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时,星标数才计入。

常见问题

AI Governance Engineer 安全吗?

它在 5 项适用的基础检查中通过了 4 项。我们没有阅读代码。在添加密钥或令牌之前,请先看看这个服务器要求哪些权限,并使用只具备所需权限的密钥。

它需要哪些密钥或代码?

注册中心没有列出必需的密钥,不过有些服务器会在首次使用时要求你登录。

它在我的电脑上运行还是在云端?

它在作者的云端运行。通过互联网连接,无需安装,但你发送的内容会经过作者的服务器。

它还在维护吗?

GitHub 仓库最后更新于前天,最新版本(0.6.1)于前天进入注册中心。

哪些应用可以使用它?

支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。它通过 URL 连接,因此也能用于接受远程 MCP 的应用,例如网页版 ChatGPT(Business 及以上为完整 MCP;Pro 为只读)。

有什么可以替代它?

同一分类中有:World Monitor, openstatus, Oh My Posh Validator, Strata, Jitsu

功能相近的其他选择

同一分类、其他作者、命令已通过检查。官方的排在前面,再按星标排序。

如何看卡片:↓ 每周下载量i★ GitHub 星标i三大支柱:评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i

World Monitor

Real-time global intelligence on conflict, markets, energy, trade, shipping and cyber. 100 tools.

作者:koala73广泛使用★ 9.1万4天前更新

在云端运行✓ 服务方 · worldmonitor.app

openstatus

Manage monitors, status pages, incidents and maintenances in your openstatus workspace.

作者:openstatusHQ广泛使用★ 92762个月前更新

在云端运行✓ 服务方 · openstatus.dev

Strata

MCP server for progressive tool usage at any scale (see https://klavis.ai)

作者:Klavis-AI正在增长★ 582414个月前更新

在云端运行仓库已归档✓ klavis.ai

Jitsu

Manage Jitsu data pipelines: destinations, streams, connections, functions, live events.

作者:jitsucom广泛使用★ 51364个月前更新

在云端运行✓ 服务方 · jitsu.com
与相似项目对比
本项
AI Governance Engineer
World MonitoropenstatusOh My Posh ValidatorStrataJitsu
运行位置在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装
需要密钥否否否否否否
可在 Claude 中按 + 添加否否否否否否
基础检查4/54/44/44/41/44/4
许可证CC-BY-4.0AGPL-3.0AGPL-3.0MITApache-2.0MIT
价格登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用

作者的 README

打开 README

按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。

AI Governance Engineer MCP server

A read-only remote Model Context Protocol server over the open data of aigovernanceengineer.com:

the obligation register, the topic crosswalk, the glossary, the patterns, the templates and

schemas, the open control profiles and the Body of Knowledge chapters. Illustrative, not legal

advice and not a claim of conformity.

The site already publishes its data as static JSON under /api/v1. This server puts an MCP layer in front of that API so an assistant can ask the questions a practitioner asks ("which EU AI Act rows bind a deployer of an Annex III system before 2028?", "what does ISO/IEC 42001 6.1.2 map to?", "give me the incident-record schema") and get answers that carry their source URL. It holds no data of its own: every answer is read from the public API (cached in memory) and every answer names the page it came from. It stores nothing about the people who call it, needs no account and writes nothing anywhere.

The public endpoint is https://mcp.aigovernanceengineer.com/mcp (Streamable HTTP, no authentication), live since 2026-09-25. You can also run it yourself (below).

Tools

Every tool is read-only (readOnlyHint: true), takes a JSON Schema input, and returns both a text answer and a structuredContent object that validates against its declared output schema. Every answer carries source (the canonical page to cite), dataset (the API document it was read from), dataVersion, license (CC BY 4.0) and notice ("Illustrative, not legal advice and not a claim of conformity"). A miss (unknown slug, id or clause) is a tool error that lists the closest valid values, so the model can correct itself.

ToolInputWhat it answers
search_glossaryquery, limit?Terms whose name, acronym or definition match every query word, best first, with the canonical /glossary/ page.
get_termslug (slug, id, URL or name)One term: definition, page, chapters that use it, other terms that mention it.
get_obligationsframework?, role?, systemClass?, appliesBefore?, status?, layer?, limit?Rows of the obligation register (obligation → artefact → stack layer) filtered with AND. framework takes an id or a name ("EU AI Act", "ISO 42001", "nist"); role matches the duty holder, or the scope of laws that name who they bind; appliesBefore keeps rows whose first application date is on or before an ISO date.
get_obligationid (e.g. AIGE-OBL-EUAIA-ART9, any case, or its URL)One row with its framework, milestones, patterns, the crosswalk references joined to it and its chapter 08 section.
map_clauseframework, refThe crosswalk topics a clause sits in and the clauses of the other frameworks on those topics, with strength (core or related), verification flag and obligation ids. Accepts "Art. 9", "Article 9(2)", "6.1.2", "GOVERN 1", "LLM01". A narrower clause falls back to its parent.
list_patternslayer? (1-5)The chapter 05 patterns, optionally for one stack layer (primary or secondary).
get_patternslug (slug, id, URL or title)One pattern: layer, summary, "Maps to", obligation ids and the full text of its page, section by section.
list_templatesstage?The templates-and-schemas library: JSON Schemas with a filled example and a human template, and the policy kit.
get_templatename, part? (all, schema, example, template)One library entry with the content of its files.
search_bokquery, kind?, limit?Chapters (title, summary, key points), sections (heading, anchor URL, text) and pattern pages, best first.
list_controlsprofile?, layer?, status?, depth?, query?The reference controls of the open control profiles (draft control specifications, ids such as AIGE-CTL-EVAL-002), filtered with AND by profile slug, stack layer (primary or secondary), publication status, depth (specified, derived, stub) and words in the id, title or objective.
get_controlid (id in any case, EVAL-002, page or JSON URL, or title)One control: objective, failure modes, scope, enforcement points, verification, evidence, failure response, patterns, the illustrative mappings, numbered references, implementation notes, open questions, the observation it records and the example observations.

Resources

The catalogue (/api/v1/index.json), the fifteen datasets (obligations, frameworks, crosswalk, glossary, patterns, maturity, path, chapters, jurisdictions, harms, cases, contracts, roles, threats, controls) and the full text (/llms-full.txt) are MCP resources under their canonical https://aigovernanceengineer.com/... URIs. Two resource templates serve a single record: https://aigovernanceengineer.com/api/v1/obligations/{id}.json (one obligation row, e.g. aige-obl-euaia-art9) and https://aigovernanceengineer.com/api/v1/controls/{id}.json (one control, e.g. aige-ctl-eval-002).

Connect a client

Claude (web and desktop). On Pro and Max plans, open *Customize > Connectors*, choose *+*, then *Add custom connector*, and paste https://mcp.aigovernanceengineer.com/mcp; leave the OAuth settings empty (the server needs no authentication). On Team and Enterprise plans an owner adds it first under *Organization settings > Connectors > Add > Custom > Web*; members then connect it from *Customize > Connectors* [5].

Claude Code.

claude mcp add --transport http aige https://mcp.aigovernanceengineer.com/mcp

Add --scope project to share it through the project's .mcp.json, or --scope user for all your projects [4]. In a JSON configuration the entry needs "type": "http" next to the URL:

{

"mcpServers": { "aige": { "type": "http", "url": "https://mcp.aigovernanceengineer.com/mcp" } } }

Other MCP clients. Point any client that supports the Streamable HTTP transport at https://mcp.aigovernanceengineer.com/mcp. The server speaks the 2026-07-28 revision of the protocol (one POST per message, no sessions) 1 and, for clients that still open with an initialize handshake (revisions 2025-03-26 to 2025-11-25), serves each request statelessly without minting a session. GET /mcp answers 405, as the 2026-07-28 revision asks of a server that has no standalone stream [2]. JSON-RPC batch arrays get 400: the protocol dropped batching in revision 2025-06-18 [9], and one POST carrying a hundred calls would count as a single request against the rate limit. Browser clients may call it directly: CORS is open, without credentials.

Configuration

Every setting comes from the environment; invalid values stop the process at start-up.

VariableDefaultMeaning
API_BASEhttps://aigovernanceengineer.com/api/v1Where the datasets are read from. Point it at a local build to test.
SITE_BASEAPI_BASE without /api/v1Where schemas, templates and llms-full.txt are read from.
PORT / HOST8787 / 0.0.0.0Listening port and interface. Use 127.0.0.1 outside a container.
PUBLIC_URLhttp://localhost: Public URL of the server (discovery document, default host allowlist).
ALLOWED_HOSTShost of PUBLIC_URL, localhost, 127.0.0.1, [::1]Host values accepted on every route but /healthz (the DNS-rebinding guard); * accepts any.
ALLOWED_ORIGINS*Browser origins allowed on /mcp (comma-separated). With a list, a request whose Origin is not on it gets 403; requests without Origin (non-browser clients) pass.
TRUST_PROXYfalseKey the rate limit on the right-most X-Forwarded-For entry. Set true only behind a proxy that sets it, such as Caddy.
TRUSTED_PROXIESemptySocket addresses of that proxy (comma-separated). With a list, X-Forwarded-For is read only on connections from one of them and anyone else is keyed by their own address; empty, any peer is trusted, which is safe only on a network nothing but the proxy can reach.
CACHE_TTL_MS3600000 (1 hour)How long a fetched document is served from memory before it is revalidated.
FETCH_TIMEOUT_MS10000Timeout of one upstream request.
MAX_UPSTREAM_BYTES16777216Largest upstream document accepted.
MAX_BODY_BYTES65536Largest request body on /mcp (413 above it).
RATE_LIMIT_MAX / RATE_LIMIT_WINDOW_MS120 / 60000Requests per client per window on /mcp (429 with Retry-After above it). A client is an IPv4 address or an IPv6 /64, since an end site holds at least a /64 and can pick any address in it [10].
MAX_CONCURRENT_REQUESTS16MCP requests handled at once; above it /mcp answers 503 with Retry-After: 1. A subscriptions/listen stream does not count.
LOG_LEVELinfodebug, info, warn or error.

How it behaves

  • Stack. TypeScript on Node.js 22, the official MCP TypeScript SDK v2 (@modelcontextprotocol/server 2.1.0) [3] with its per-request createMcpHandler, and Hono for the HTTP routes.
  • Data. Each dataset is fetched from API_BASE on first use and kept in memory. Within the TTL it is served from memory; after it, the next reader revalidates with If-None-Match and If-Modified-Since, and a 304 extends the copy without a download. Concurrent readers share one request. If the site cannot be reached, the last copy keeps being served, the failure is logged and the site is asked again after a minute (or the TTL, if shorter), not on every tool call; with no copy, the tool answers with an error that names the URL. The chapter text and the pattern pages come from /llms-full.txt, split into documents and headings; section anchors are recomputed the way the site renders them, so the URLs point at the right heading.
  • Templates list. The site publishes no index of its templates and schemas, so the list is bundled in src/catalogue.generated.ts (from site/public/schemas, site/public/templates and site/src/data/templates.ts); the file contents are still fetched live. A schema published later is served by name anyway (/schemas/ .v1.json is tried). Run npm run catalogue after adding a schema or template; test/resources.test.ts fails until you do.
  • Privacy. Logs are JSON lines: method, path, status, duration and, when the client sends them, the Mcp-Protocol-Version, Mcp-Method and Mcp-Name headers. The last two are logged only when they hold a known method or one of this server's tool names; a resource address or any other text is written as [other]. For a POST to /mcp they add, from the body, the JSON-RPC method (rpc), the tool a tools/call names (tool) and the name and version the client announces in initialize (client, clientVersion). Each is checked against a list of known values or a strict pattern and written as [other] if it fails; nothing else from the body is read into the log. A client can mark its own traffic with ?tag= on the URL (lowercase letters, digits and hyphens, up to 24 characters; logged as tag), so an operator can leave their own use out when counting. The tag is not a secret: anyone can send it, so a count that leaves a tag out is a lower bound. The client address, the user agent and what was asked (tool arguments, search text, the resource read) are never logged. The rate limiter keys clients by an HMAC of their address with a secret drawn at start-up, drops each entry when its window closes and forgets everything on restart. A client past its limit is logged once per window, not on every refused request.
  • Stateless. No sessions, no storage, no background jobs. Any number of replicas can run side by side; each keeps its own cache and its own rate-limit table.
  • Routes. POST /mcp (MCP)

只显示了开头部分;其余内容请在仓库中阅读