Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
作者提供的说明(英文原文)
Create, edit, validate, render, and export D2 diagrams with confined workspace access.
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。
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。
选择你使用的 AI 应用,然后复制设置。
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": {
"d2mcp": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/recursivefunctions/d2mcp:0.5.0"
]
}
}
}在终端中运行这条命令
claude mcp add d2mcp -- docker run -i --rm ghcr.io/recursivefunctions/d2mcp:0.5.0粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"d2mcp": {
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/recursivefunctions/d2mcp:0.5.0"
]
}
}
}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。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"d2mcp": {
"type": "stdio",
"command": "docker",
"args": [
"run",
"-i",
"--rm",
"ghcr.io/recursivefunctions/d2mcp:0.5.0"
]
}
}
}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。
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6日
注册名称已通过 GitHub 账户 recursivefunctions 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
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。
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。
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)。
GitHub 仓库最后更新于–,最新版本(0.5.0)于40天前进入注册中心。
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。
同一分类中有:Spec Driven Development, Remote Desktop Commander, debuff
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
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Let coding agents play, test and tune your 2D web game: live state, explained errors, replays, fuzz.
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。
A Model Context Protocol (MCP) server that provides D2 diagram generation and manipulation capabilities.
D2 is a modern diagram scripting language that turns text to diagrams. This MCP server allows AI assistants like Claude to create, render, export, and save D2 diagrams programmatically.
The server provides 16 MCP tools for complete-source authoring, validation, rendering, embedded D2 guidance, and incremental Oracle editing.
With the new Oracle API integration, AI assistants can now build and modify diagrams incrementally, making it perfect for:
d2mcp/
├── cmd/ # Application entry point ├── internal/ │ ├── domain/ # Business entities and interfaces │ │ ├── entity/ # Domain entities │ │ └── repository/ # Repository interfaces │ ├── usecase/ # Business logic │ ├── infrastructure/ # External implementations │ │ ├── d2/ # D2 library integration │ │ └── mcp/ # MCP server implementation │ └── presentation/ # MCP handlers │ └── handler/ # Tool handlers └── pkg/ # Public packages
All export formats use bundled Go renderers; no rsvg-convert, ImageMagick, or Chromium installation is required.
The embedded capability catalog is based on the official D2 language tour, D2 API documentation, and D2 v0.9.0 release.
# Clone the repository
git clone https://github.com/recursivefunctions/d2mcp.git cd d2mcp
make build
make build-all
go install github.com/recursivefunctions/d2mcp/cmd@latest
docker run --rm -i \
--mount type=bind,src="$PWD",dst=/workspace \ ghcr.io/recursivefunctions/d2mcp:0.5.0
The image defaults to STDIO transport and confines local imports, assets, and exports to the mounted /workspace directory. The container runs as a non-root user, so the mounted directory must be writable by that user when saving exports.
Release tags publish the server as io.github.RecursiveFunctions/d2mcp in the official MCP Registry. Registry publication and editor-gallery indexing are separate, so a newly published version might not appear immediately in an editor's search results.
Compatible clients can also use the portable .mcp.json configuration in this repository. It requires Docker and mounts the current workspace at /workspace.
Client configurations are generated from server.json, so they always use the same immutable container version.
| Client | Installation |
|---|---|
| VS Code | Use the installation URL, Official MCP Registry, or companion extension |
| VSCodium and Positron | Install the RecursiveFunctions.d2mcp companion extension from Open VSX |
| Cursor | Use the generated MCP configuration |
| Windsurf | Use the generated MCP configuration |
| Trae | Use the generated install URL or project configuration |
| Antigravity | Use the generated MCP configuration |
| Zed | Use the generated manual configuration while Zed completes Official Registry integration |
| Kilo Code | Install from Kilo Marketplace after its catalog PR is accepted, or use the portable configuration |
| Cline | Install from Cline Marketplace after curator approval, or follow the Docker instructions above |
Run make generate-adapters after changing release metadata. CI runs make verify-adapters and rejects stale client configurations.
# Simple build
make build
make run
make build-all
Add to your Claude Desktop configuration file:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json Linux: ~/.config/Claude/claude_desktop_config.json
For STDIO transport (recommended for Claude Desktop):
{
"mcpServers": { "d2mcp": { "command": "/path/to/d2mcp", "args": ["-transport=stdio"] } } }
For SSE transport:
{
"mcpServers": { "d2mcp": { "command": "/path/to/d2mcp", "args": ["-transport=sse", "-addr=:3000"] } } }
Replace /path/to/d2mcp with the actual path to your built binary.
# Run the MCP server (stdio transport)
./d2mcp -transport=stdio
./d2mcp
./d2mcp -transport=sse
./d2mcp -transport=streamable
d2mcp now supports multiple transport protocols:
The traditional stdio transport for direct process communication:
./d2mcp -transport=stdio
HTTP-based transport that allows network connectivity:
# Basic SSE mode (defaults to :3000)
./d2mcp -transport=sse
./d2mcp -transport=sse \ -addr=:8080 \ -base-url=http://localhost:8080 \ -base-path=/mcp \ -keep-alive=30
SSE Configuration Options:
-addr: Address to listen on (default: ":3000")-base-url: Base URL for SSE endpoints (auto-generated if not specified)-base-path: Base path for SSE endpoints (default: "/mcp")-keep-alive: Keep-alive interval in seconds (default: 30)SSE Endpoints: When running in SSE mode, the following endpoints are available:
http://localhost:3000/mcp/ssehttp://localhost:3000/mcp/messageThe modern HTTP-based transport that simplifies bidirectional communication:
# Basic Streamable HTTP mode
./d2mcp -transport=streamable
./d2mcp -transport=streamable \ -addr=:8080 \ -endpoint-path=/mcp \ -heartbeat-interval=30 \ -stateless
Streamable HTTP Configuration Options:
-addr: Address to listen on (default: ":3000")-endpoint-path: Endpoint path for Streamable HTTP (default: "/mcp")-heartbeat-interval: Heartbeat interval in seconds (default: 30)-stateless: Enable stateless mode (default: false)Streamable HTTP Endpoint: When running in Streamable HTTP mode, a single endpoint handles all communication:
http://localhost:3000/mcpThe server confines imports, local images, and saved files to named workspace roots. The default root is the server's current working directory.
./d2mcp -transport=stdio \
-workspace-root=docs=/workspace/docs \ -workspace-root=diagrams=/workspace/diagrams \ -remote-asset-max-bytes=10485760 \ -remote-asset-timeout=30s
-workspace-root is repeatable. Local path traversal and symlink escapes are rejected. Remote assets require HTTPS and are subject to DNS/IP, redirect, timeout, MIME, and response-size checks; loopback and private-network destinations are rejected.
Create a new diagram with optional initial content (unified approach):
Empty diagram (for Oracle API workflow):
{
"id": "my-diagram" }
With initial D2 content and a named workspace for imports/assets:
{
"id": "my-diagram", "content": "a -> b: Hello\nserver: {shape: cylinder}", "workspace_root": "diagrams" }
Export a diagram using a bundled renderer and layout engine:
{
"diagramId": "my-diagram", "format": "png", "layout": "tala", "themeId": 200, "scale": 1.5, "maxPixels": 33554432 }
Formats are svg, png, pdf, pptx, gif, and ascii (txt is an alias). Layouts are dagre, elk, and tala. Typed options cover engine spacing, themes and overrides, bundled fonts, padding, scale, sketch mode, animation, board selection, and resource limits.
PDF, PPTX, GIF, and optionally animated SVG compose bounded descendant boards. PNG and ASCII produce one board; use boardPath to select one from a composition.
Save a diagram inside a configured workspace root:
{
"diagramId": "my-diagram", "format": "pdf", "workspace_root": "diagrams", "path": "exports/output.pdf" }
If path is omitted, output is written below d2mcp_output/ in the selected root.
Validate raw D2 source or the current source of a stored diagram. Supply exactly one input:
{
"content": "a: {\n b" }
{
"diagram_id": "my-diagram" }
Invalid D2 returns a successful structured result with valid: false and compiler diagnostics. Simple, unambiguous terminator errors may also include repaired_content; validation never modifies a stored diagram.
Complete D2 source is canonical, so every bundled v0.9 language feature can be authored without waiting for a specialized structured tool.
{"diagram_id": "my-diagram"}
{
"diagram_id": "my-diagram", "content": "direction: right\nclient -> api -> database" }
Updates compile first and replace atomically. Invalid drafts return structured diagnostics and leave stored source unchanged.
{"content": "api:{db}"}
Returns formatted source without creating or changing a diagram.
{"query": "sequence diagrams", "limit": 5}
d2_help searches the embedded v0.9.0 reference or returns an exact topic by ID. d2_capabilities lists its categories and MCP resource URIs. Embedded resources cover the core language, connections, reusable definitions, visuals, rich text, structured/sequence/grid diagrams, composition, layouts, and exports.
The Ora
只显示了开头部分;其余内容请在仓库中阅读