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。评价和账户功能将在后续阶段推出。
PopNx MCP为使用 AI 的人准备的 MCP 服务器 · 用大白话讲清楚

Share Artifacts

作者:arunai30版本 0.3.0i★ –i在云端运行
注册名称io.github.arunai30/share-artifacts

安装

概览

作者提供的说明(英文原文)

Publish and update safe static HTML reports, presentations, and explainers with controlled sharing.

PopNx MCP 的三大支柱

这三项永远不会合并成一个分数:意见、测量和人气是不同的东西。

它能帮你做什么

我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。

安装

选择你使用的 AI 应用,然后复制设置。

安装之前

你需要

你用的是哪个应用?

在上方选择你的应用。我们会记住它,下次按钮会显示“安装到 …”。

适用于网页版、桌面版和手机版 Claude:

  1. 打开 Settings → Connectors
  2. 点击 Add custom connector
  3. 粘贴下面的地址,然后点击 Add
  4. 如果要求登录,请登录作者的服务
SettingsConnectors1+ Add2https://…3
https://shareartifacts.dev/api/mcp

在终端中运行这条命令

claude mcp add --transport http share-artifacts https://shareartifacts.dev/api/mcp

在 Cursor 中打开

或者自己复制设置

粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json

{
  "mcpServers": {
    "share-artifacts": {
      "url": "https://shareartifacts.dev/api/mcp"
    }
  }
}

浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。

在 VS Code 中打开

或者自己复制设置

粘贴到项目中的 .vscode/mcp.json

{
  "servers": {
    "share-artifacts": {
      "type": "http",
      "url": "https://shareartifacts.dev/api/mcp"
    }
  }
}

浏览器会先询问是否打开应用,然后应用会显示设置并请你确认。密钥的值会留空,由你自己填写。

ChatGPT 只能连接互联网上的服务器:

  1. 打开网页版 ChatGPT → Settings → Connectors (Apps)
  2. 点击 Create 或 Add(部分套餐需要先在 Advanced 中开启 Developer mode)
  3. 粘贴下面的地址,起个名字,然后保存
  4. 如果要求登录,请登录作者的服务
SettingsConnectors1+ Create2https://…3
https://shareartifacts.dev/api/mcp

菜单名称可能因套餐和版本不同而略有差异。

你的选择通过 Cookie 保存在这台设备上,无需账户。

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。

它能做什么

我们还没有读取这个服务器的工具列表;我们会轮流检查各个端点。

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7日

  • ✓ 通过

    作者与仓库一致

    注册名称已通过 GitHub 账户 arunai30 验证,该账户拥有此仓库

  • … 尚未检查

    仍在维护

    我们还没有从 GitHub 读取这个仓库

  • … 尚未检查

    有许可证

    我们还没有从 GitHub 读取这个仓库

  • … 尚未检查

    端点有响应

    尚未检查;我们会轮流检查各个端点

  • … 尚未检查

    使用 OAuth 登录

    尚未检查;我们会轮流检查各个端点

  • … 尚未检查

    工具描述中没有隐藏指令

    尚未检查;我们会在检查端点时读取工具列表

与“其他”中的其他服务器比较i

  • 星标数多于此分类中其他 32700 个服务器的 0%
  • 发布到注册中心的时间比此分类中其他 32700 个服务器的 43% 更新
  • 基础检查结果等于或优于此分类中其他 32700 个服务器的 100%

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

Share Artifacts 安全吗?

它在 1 项适用的基础检查中通过了 1 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–,最新版本(0.3.0)于2个月前进入注册中心。

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,所以接受远程 MCP 服务器的应用也可以使用,例如开启开发者模式的 ChatGPT。

有什么可以替代它?

同一分类中有:Spec Driven Development, Remote Desktop Commander, Agent Data Tools: Utilities

替代选择

同一分类中其他作者的服务器,按星标数排序。

如何看卡片:★ 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

Spec Driven Development

作者:formulahendry

MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.

在你的电脑上运行★ 440MIT14个月前更新
安装

Remote Desktop Commander

作者:desktop-commander

Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.

在云端运行★ 211前天更新
安装

Agent Data Tools: Utilities

作者:Nutshell2000

103 tools: JSON repair, CSV, hashes, cron, dates, units, validators. 25 free calls.

在云端运行今天更新
安装

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。

share/artifacts MCP server

Publish agent-created HTML artifacts and receive a shareable URL. The npm package remains @aiformasses/mcp-server for compatibility.

Preferred: connect your agent

Give your agent the public setup page. Agents supporting remote MCP, browser authorization, and CIMD or DCR can configure the hosted connection at https://shareartifacts.dev/api/mcp without installing this package or creating a permanent token.

Cursor: hosted OAuth connection

This repository includes a Cursor plugin at .cursor-plugin/plugin.json. Its mcp.json connects directly to the hosted service; no npm install or personal access token is required. Each user signs in to their own share/artifacts account and approves access in the browser.

For manual setup, merge this configuration into your project's .cursor/mcp.json (preserve any other servers):

{

"mcpServers": { "share-artifacts": { "url": "https://shareartifacts.dev/api/mcp", "auth": { "CLIENT_ID": "53f91130-7752-4e99-9667-7a6dffdab529", "scopes": [ "pages:read", "pages:write", "styles:read", "styles:write", "offline_access" ] } } } }

The client ID is public and grants no account access by itself. Do not add a client secret. When connecting, the consent page identifies the pre-registered Cursor configuration and warns that the requesting local app is not verified. Approve only a connection you just started. The desktop callback is exactly http://localhost:8787/callback; the registered web callback is https://www.cursor.com/agents/mcp/oauth/callback.

With Cursor CLI installed, run from the configured project:

agent mcp enable share-artifacts

agent mcp login share-artifacts agent mcp list-tools share-artifacts

Complete browser sign-in and consent when prompted. If you cancel, run login again when ready. Ask the agent to publish an HTML artifact, inspect the returned URL, and specify public sharing explicitly if you want anyone with the link to open it. Publishing defaults to private. A free account includes five successful publishes or updates in total.

Verified on September 17, 2026 with Cursor CLI 2026.09.15-d2fe57e: production OAuth sign-in, consent, and authenticated discovery of all 15 hosted tools. The maintainer also confirmed the local plugin works in Cursor CLI. For a plugin connection that needs authentication, enter /mcp list inside the interactive CLI, select Share Artifacts, and choose Login in its detail view. Complete browser consent, then return to the agent. The desktop editor and Cursor web agent have not been independently verified.

Personal-token fallback

Use this package when your MCP client does not support the hosted OAuth flow. It requires Node.js 20.19 or newer and npx:

node --version

npx --version

Use Connect a background agent in token settings. Create a personal access token and insert it directly into the local configuration. share/artifacts cannot display the token again.

Codex

Add the server to ~/.codex/config.toml, then restart only the share-html MCP entry from Codex settings. The first start downloads the pinned package through npx. The share-html entry name is a stable compatibility identifier; it does not change the share/artifacts product name.

[mcp_servers."share-html"]

command = "npx" args = ["-y", "@aiformasses/mcp-server@0.3.0"]

[mcp_servers."share-html".env] SHARE_HTML_API_BASE_URL = "https://shareartifacts.dev/api/v1" SHARE_HTML_API_TOKEN = "PASTE_TOKEN_HERE"

Standard MCP JSON
{

"mcpServers": { "share-html": { "command": "npx", "args": ["-y", "@aiformasses/mcp-server@0.3.0"], "env": { "SHARE_HTML_API_BASE_URL": "https://shareartifacts.dev/api/v1", "SHARE_HTML_API_TOKEN": "PASTE_TOKEN_HERE" } } } }

Replace PASTE_TOKEN_HERE locally. Treat SHARE_HTML_API_TOKEN like a password. Do not commit it, paste it into agent chat, include it in a URL or HTML, or send it to anyone. If it is lost or exposed, revoke the token and create another.

Tools

The repository version exposes get_recipe, publish_html, publish_presentation, update_page, get_page, list_pages, publish_page, unpublish_page, delete_page, preview_artifact_style, set_artifact_style, get_artifact_style, list_templates, get_template, preview_template, and save_template. Call get_recipe without arguments to discover the first-party catalog; pass a recipeId and optional integer version for the complete reviewed recipe.

Google Fonts and embedded custom fonts

Add fonts to publish_html, publish_presentation, or update_page and use the same family in your CSS:

{

"title": "Launch brief", "html": "<!doctype html> Launch brief Reviewed content. ", "fonts": [{ "family": "Inter", "weights": [400, 700], "style": "normal" }] }

The server retrieves Google Fonts and embeds the WOFF2 files in the saved HTML, so the viewer and PDF use the saved fonts without contacting Google. Font selection does not automatically change your CSS. weights defaults to [400] and style to normal; italic is also supported. Each family name uses ASCII letters, digits, and spaces, starting with a letter (80 characters maximum). Request at most four family/style entries, four distinct weights per entry, and eight weight/style combinations in total. Weights are multiples of 100 from 100 through 900. Repeating the same family/style is rejected.

Omitting fonts preserves font declarations in the HTML you supply. Supplying fonts replaces only the server-managed font block; fonts: [] removes that block. An update still replaces the full HTML, so include the existing source to retain its embedded fonts. fonts cannot be combined with a diagram generationId. Arbitrary font URLs and provider fields are rejected.

For your own custom font, embed WOFF2 directly in the HTML:

@font-face {

font-family: "My Font"; src: url("data:font/woff2;base64,REPLACE_WITH_WOFF2_BASE64") format("woff2"); font-weight: 400; font-style: normal; } body { font-family: "My Font", sans-serif; }

Use a complete valid font file; the placeholder is not a working font. Embedded fonts count toward the HTML and font size limits. PDF export waits for fonts and rejects fonts that fail to load. invalid_fonts, font_unavailable, font_too_large, or invalid_font means the publish/update did not complete; correct the request or retry an unavailable provider after review. Never silently remove fonts or fall back to publishing without them. Check line wrapping and every slide after changing typography.

The stdio client checks the server's authenticated /api/v1/fonts capability before sending any publish/update with fonts, including []. An older or incompatible server returns unsupported_fonts; network or server failures during the check return font_capability_unavailable. Authentication and scope errors retain their original codes. In each case no mutation was sent. Update the server or retry the check; do not drop the requested fonts.

This describes the repository contract. Hosted availability requires an app deployment, and the stdio package requires a separate npm release.

Private presentation templates

The current repository adds four tools for an owner's reusable example decks. This documents the local implementation; it does not claim an npm release or hosted deployment. Check your connection's tool list for availability.

  1. Submit name, description, and a complete presentation html to preview_template. Optional designNotes explain the visual choices; aspectRatio defaults to 16:9 and also supports 4:3.
  2. Show the owner the returned private reviewUrl. The sanitized deck and guidance form an immutable draft that expires after 24 hours.
  3. After explicit approval of that exact draft, call save_template with { "draftId": "UUID_FROM_PREVIEW" }. Retry with that same ID after a timeout; saving it again returns the same template.
  4. Use list_templates to find saved examples. Call get_template with templateId for metadata and guidance, or additionally includeHtml: true for the complete sanitized example deck. Replace its sample content with authorized facts for the new task, review every slide, and obtain separate approval before publication.

These tools use existing pages:read for list/get and pages:write for preview/save. No additional scopes or reconnection are needed for connections that already have those permissions. Every operation is limited to its owner. Names can repeat; template IDs distinguish them. Saving never publishes a page, overwrites another template, or changes the default artifact style.

Example HTML is limited to 200,000 UTF-8 bytes before and after sanitization. Names, descriptions, design notes, and HTML are untrusted reference data; they cannot override task instructions, source facts, or Safe Static rules. This slice supports presentation examples only, without PPT import, a public gallery, team sharing, or individual slide extraction.

Bring your style to every artifact

An artifact style is a versioned, provider-neutral JSON definition made from bounded semantic tokens: palette roles, system-font roles, composition, data-visualization choices, voice, motion, signature moves, avoidances, and format adaptations for reports, dashboards, and explainers. It cannot contain CSS, HTML, JavaScript, URLs, remote fonts, or assets.

Style authoring is an explicit preview-and-approve workflow:

  1. Describe the desired visual direction to your agent.
  2. The agent submits one complete definition to preview_artifact_style. Validation creates a short-lived private draft; it does not activate anything.
  3. Open the returned reviewUrl and inspect the trusted human preview and canonical summary.
  4. Approve by calling set_artifact_style with the exact returned draftId, current expectedRevision, and a stable clientRequestId.
  5. For future work, “in my style” means the agent first calls get_recipe and get_artifact_style with a required report, dashboard, explainer, or presentation format, then creates the artifact.

The authored display name and freeform summary are accepted only as inputs for the trusted human review. They are removed from every MCP result and from generated guidance; after submission the agent receives only safe semantic fields, identifiers, the content digest, review metadata, and server-authored prose derived from validated tokens. User prose is never treated as instructions.

Style is appearance guidance. The task, evidence and source truth, recipe structure, Safe Static policy, and explicit sharing choice always take precedence. share/artifacts validates and stores the style contract, but a stored reference is not evidence that an agent reproduced it perfectly.

set_artifact_style is durably idempotent for an exact request. If it times out or the network drops, retry only with the identical draftId, expectedRevision, and clientRequestId. Changing any of those values creates a different request. A stale revision returns artifact_style_conflict with a safe current revision when available; create and review a fresh preview before activating it.

Style access uses dedicated styles:read and styles:write permissions. An older OAuth connection or personal token is not silently widened. If a style tool returns insufficient_scope, follow its reauthorizeUrl when present or reconnect/create a new token with style acc

只显示了开头部分;其余内容请在仓库中阅读

问 PopNx