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

Atelier

作者:marmikshah版本 0.2.0i★ –i在你的电脑上运行
注册名称io.github.marmikshah/atelier

安装

概览

作者提供的说明(英文原文)

Offline, headless pixel-art editor: draw, animate, inspect, and export sprites and fonts.

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。

  • 连接方式: 在你的电脑上运行
  • 传输方式: stdio
  • 分类: 其他

安装

选择你使用的 AI 应用,然后复制设置。

安装之前

  • 基础检查:1 项中通过 1 项
  • 在你的电脑上运行,拥有你的用户账户的权限(文件、网络、应用)。
  • 注册中心没有列出需要填写的密钥。

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 …”。

打开 Claude Desktop → Settings → Developer → Edit Config,把下面的内容粘贴到 claude_desktop_config.json

{
  "mcpServers": {
    "atelier": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/marmikshah/atelier:0.2.0"
      ]
    }
  }
}

在终端中运行这条命令

claude mcp add atelier -- docker run -i --rm ghcr.io/marmikshah/atelier:0.2.0

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": {
    "atelier": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/marmikshah/atelier:0.2.0"
      ]
    }
  }
}

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": {
    "atelier": {
      "type": "stdio",
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "ghcr.io/marmikshah/atelier:0.2.0"
      ]
    }
  }
}

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。

这个服务器在你的电脑上运行,而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。

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

它在你的电脑上运行;安装后,工具列表会显示在你的 AI 应用中。

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 marmikshah 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

  • – 不适用

    端点有响应

    它在你的电脑上运行,没有互联网端点

  • – 不适用

    使用 OAuth 登录

    在你的电脑上运行,改用本地设置

  • … 尚未检查

    工具描述中没有隐藏指令

    它在你的电脑上运行;我们只能读取互联网端点的工具列表

与“其他”中的其他服务器比较i

  • 星标数多于此分类中其他 32691 个服务器的 0%
  • 发布到注册中心的时间比此分类中其他 32691 个服务器的 66% 更新
  • 基础检查结果等于或优于此分类中其他 32691 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

Atelier 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?

它在你的电脑上运行。你的 AI 应用会在需要时启动它,数据不会经过作者的服务器(除非它自己调用在线服务)。

它还在维护吗?

GitHub 仓库最后更新于–,最新版本(0.2.0)于19天前进入注册中心。

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。你的电脑上还需要有 Docker。

有什么可以替代它?

同一分类中有:Spec Driven Development, Remote Desktop Commander, Dialyne

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

Dialyne

作者:dialyne

Real phone calls in France for AI agents: dial +33 numbers, menus, hold, AI disclosure.

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。

Atelier is an offline, headless pixel-art editor for command-line tools and AI agents. Create layered sprites and animations, inspect pixels, and export PNGs, spritesheets, GIFs, APNGs, and TrueType pixel fonts. Documents stay local.

Use it directly from a shell, or connect an agent such as Claude Code or Codex through MCP (Model Context Protocol). Both interfaces expose the same editing tools; no graphical editor or running server is needed for CLI calls.

Drawn by an agent through Atelier's tools. Each is rebuilt byte for byte from the recipe beside it in CI.

A more complex design made by combining physics simulation with Atelier MCP using Opus 5.5.

The replay file is too large to include here. Download it from Google Drive if needed.

Showcase

Install

On macOS, Ubuntu, or any other Linux, paste this into a terminal:

curl -fsSL https://github.com/marmikshah/atelier/releases/latest/download/install.sh | sh

Ubuntu Desktop comes without curl; use wget instead:

wget -qO- https://github.com/marmikshah/atelier/releases/latest/download/install.sh | sh

It downloads Atelier for your machine, checks it, puts it in ~/.local/bin, and adds that to your PATH. If Claude Code or Codex is installed, it also connects Atelier to them: they get its drawing tools in every project and its guides for drawing sprites and scenes and reviewing art. Nothing else is needed, not even Rust.

Then start a new agent session and ask for something:

Use atelier to draw a 32×32 animated torch and save it as torch.gif.

Your art is kept in ~/.atelier. To keep a project's art inside the project instead, run atelier init in its folder before you start the agent.

  • Update: Run the install command again.
  • Installed Claude Code or Codex afterwards? Run atelier connect.
  • A specific version or folder, or no profile changes: Set ATELIER_VERSION, ATELIER_INSTALL_DIR, or ATELIER_NO_MODIFY_PATH=1; install.sh lists every option.

Uninstall

atelier uninstall

It lists what it will remove and asks before deleting anything: Atelier's registration in Claude Code and Codex, its skills, the PATH line it added, the atelier program, and the Linux background server if you set one up.

Your art in ~/.atelier is kept; add --documents to delete it as well. Folders set up with atelier init are never touched.

Other ways to run it

  • From the terminal. Every tool is one command: atelier call TOOL ' '. Your first sprite walks through creating, drawing, and saving an image.
  • In Cursor and other MCP clients. Install as above, then register the path command -v atelier prints as a stdio server with no arguments, and run atelier skills install --for cursor (or --dir PATH for another skills folder) for the drawing guides:
  {

"mcpServers": { "atelier": { "command": "/full/path/to/atelier" } } }

  • In Docker. A container image is published for MCP clients that can launch Docker.
  • As an HTTP server or background service. See Running Atelier.
Build from source

On other platforms, or to build it yourself, install Rust 1.98 or later, then:

cargo install --locked --git https://github.com/marmikshah/atelier atelier

atelier connect

atelier uninstall removes a build installed this way too, through cargo uninstall.

I want to…Start here
Make art from the terminalGetting started
Run a server, use Docker, or choose where art is storedRunning Atelier
Find a tool's argumentsatelier tools --markdown or atelier tools --schema NAME
Build from a checkout, understand the code, or work on the showcaseDevelopment guide
Report a bug or security issueIssues · Security

Stability

Atelier is at 0.2. master is the current source; releases are dispatched from it and publish the macOS and Linux binaries, the container image, a vX.Y.Z tag, and the MCP Registry entry.

  • Your documents are not discarded. Edits are journaled and committed atomically. A document that is damaged, or saved by a build with a different document schema, is refused with an error and left unchanged on disk.
  • The document format is versioned, with no migration. If an update changes it, documents saved before it are refused, stay on disk untouched, and open again with the commit that wrote them.
  • A recipe is the plain log of the calls that built a document, one JSON line each. It has no version of its own: what it draws depends on how the tools behave, so a recipe from one commit replays exactly on commits whose tools draw the same way.
  • Tool names and arguments can still change between commits. Pin a commit if you script against them, and read the current surface from atelier tools --markdown.

How it is built

100% of this code was written by AI. Atelier began as an experiment in whether agents, using only tool calls, can make art useful in games. My part has been giving direction, and I hope the result helps you make something of your own.

CI checks formatting, strict Clippy, and the Rust test suite on Linux and macOS, smoke-tests the container, and replays every showcase recipe, which must reproduce its GIF byte for byte. The security policy describes the network and file-access boundaries.

[!NOTE]

None of the code has had line-by-line human review. Bugs, security issues, and breaking changes are still possible. Please review the code and work with copies of important data before using it in production.

Contributing

Bug reports and questions are welcome as issues; include the checkout commit and a minimal command sequence or recipe file. Pull requests from outside the project are not accepted. Report vulnerabilities privately through the security policy.

License

MIT © Marmik Shah

问 PopNx