
Markmap MCP Server
MCP把 Markdown 大纲转换成可交互的思维导图 HTML,也能在服务端导出 PNG / JPG / SVG(需安装 Playwright)。完全本地运行,无需第三方 API Key,适合整理笔记、备课和梳理文档结构。
Markmap MCP Server
Markmap MCP Server is based on the Model Context Protocol (MCP). It converts Markdown into interactive mind maps using markmap, and can optionally export PNG / JPG / SVG on the server for Agent-friendly delivery. Generation runs locally — no third-party API keys required.
Features
- Markdown → Mind Map: Convert Markdown (headings + nested lists) to interactive HTML mind maps
- Agent-friendly returns: Return a file path, inline HTML, and/or image content (configured at startup)
- Server-side export: Export PNG / JPG / SVG via Playwright for chat/inline preview
- Browser preview: Configurable open behavior — always, never, or agent-decided (startup setting)
- Browser export toolbar: When viewing HTML, also export PNG/JPG/SVG or copy Markdown in the page UI
- Offline HTML: Startup
--offlineinlines assets so the page works without CDN access - File workflows: Read from
inputPath, list recent outputs, clean up old files - Privacy-first: Fully local conversion; no cloud mind-map API
Prerequisites
- Node.js v20 or above
- For server-side image export (
format: png|jpg|svg): install Playwright and Chromium
npm install playwright
npx playwright install chromium
(playwright is an optional dependency of this package.)
Installation
# Install from npm
npm install @jinzcdev/markmap-mcp-server -g
# Basic run
npx -y @jinzcdev/markmap-mcp-server
# Specify output directory and auto-open in browser
npx -y @jinzcdev/markmap-mcp-server --output /path/to/output/directory --open always
Docker
docker build -t markmap-mcp-server .
docker run --rm -i \
-v /path/to/output:/data/markmap \
-e MARKMAP_DIR=/data/markmap \
markmap-mcp-server
Or clone and run locally:
git clone https://github.com/jinzcdev/markmap-mcp-server.git
cd markmap-mcp-server
npm install && npm run build
# Optional: enable server-side image export
npx playwright install chromium
node build/index.js
Usage
Add the following configuration to your MCP client (Cursor / Claude Desktop / etc.):
{
"mcpServers": {
"markmap": {
"type": "stdio",
"command": "npx",
"args": ["-y", "@jinzcdev/markmap-mcp-server"],
"env": {
"MARKMAP_DIR": "/path/to/output/directory",
"MARKMAP_OPEN": "never",
"MARKMAP_RETURN_MODE": "path"
}
}
}
}
Server preferences (CLI / env)
These are decided when the server starts — not tool arguments:
| Preference | CLI | Env | Values | Default |
|---|---|---|---|---|
| Output directory | --output / -o | MARKMAP_DIR | any directory path | ~/.markmap-mcp |
| Open in browser | --open [mode] | MARKMAP_OPEN | always | never | agent (bare --open = always) | never |
| Return mode | --return-mode | MARKMAP_RETURN_MODE | path | content | both | path |
| Offline HTML | --offline | MARKMAP_OFFLINE | true | false (CLI: pass --offline to enable) | false |
CLI flags override environment variables. --output overrides MARKMAP_DIR.
--open: bare --open means always. You may also pass --open always|never|agent. An invalid value exits with an error. Without the flag, MARKMAP_OPEN is used (default never).
Return mode:
| Mode | Meaning |
|---|---|
path | Paths JSON only (htmlFilePath + filePath) |
content | Inline content only (raw HTML text, or a base64 image block) — no paths |
both | Paths JSON plus inline content |
Generated HTML always includes the markmap toolbar, English export labels, and fully expanded nodes.
Example prompts
- “Summarize this design doc as a mind map.”
- “Convert
./notes/architecture.mdto a mind map.” - “Generate a PNG mind map of this outline for the chat.”
Available Tools
markdown_to_mindmap
Convert Markdown into an interactive mind map (and optionally an image).
| Parameter | Type | Default | Description |
|---|---|---|---|
markdown | string | — | Markdown content (required unless inputPath is set). Wins when both are provided. |
inputPath | string | — | Absolute path to a local .md file |
format | html | png | svg | jpg | html | Output format. Image formats need Playwright |
filename | string | auto | Output base name (sanitized; prefixed with markmap- if needed; reusing overwrites) |
open | boolean | false | Whether to open the result in browser. Only exposed when server open mode is agent (--open agent / MARKMAP_OPEN=agent). |
Return (returnMode=path):
{
"htmlFilePath": "/path/to/markmap-….html",
"filePath": "/path/to/markmap-….html"
}
For image formats, filePath is the image path and htmlFilePath is still the HTML source.
Return (returnMode=content): Raw HTML text block, or an MCP image content block (base64) for png/jpg/svg — no paths JSON. Oversized HTML (≥200KB) falls back to paths JSON.
Return (returnMode=both): Paths JSON plus the inline content above.
Note: Interactive zoom/collapse and the in-page “Export PNG/JPG/SVG” buttons are part of the HTML viewer. Server-side
format: png|jpg|svgis what Agents can consume directly without a manual browser click.
list_mindmaps
List recent generated files in the output directory (newest first). Only files whose names start with markmap are included.
| Parameter | Type | Default | Description |
|---|---|---|---|
limit | number | 20 | Max files to return (1–200) |
Returns {outputDir, files: [{name, filePath, size, mtimeMs, mtime}]}.
get_mindmap
Retrieve a generated mind map file by its absolute path. Path must stay inside the configured output directory (path traversal denied).
| Parameter | Type | Default | Description |
|---|---|---|---|
filePath | string | — | Absolute path to the mind map file (HTML or image) |
Returns JSON {filePath, mimeType, size}. For HTML/SVG under 200KB, also appends a text content block. PNG/JPG return metadata only (no image block) — re-export via markdown_to_mindmap with format=png\|jpg if pixels are needed.
cleanup_mindmaps
Delete old (or all) generated mind map files.
| Parameter | Type | Default | Description |
|---|---|---|---|
maxAgeDays | number | 7 | Delete files older than this many days |
all | boolean | false | If true, delete all markmap files |
dryRun | boolean | false | If true, preview without deleting |
Prompt: mindmap_from_content
Helper prompt that asks the model to structure notes as Markdown, then call markdown_to_mindmap.
| Parameter | Type | Default | Description |
|---|---|---|---|
topic | string | — | Topic or raw notes to organize as a map |
Related Projects
| Project | Description |
|---|---|
| MarkXMind Online | Create XMind mind maps with Markdown online. Try it → |
| Obsidian MarkXMind Plugin | Render XMindMark syntax as XMind mind maps inside Obsidian. |
License
This project is licensed under the MIT License.
相關資源

markmap-cli
CLI把 Markdown 大纲转成可折叠、可缩放的交互式 HTML 思维导图,支持离线内联资源与监听文件实时刷新,适合梳理需求、备课与整理知识结构。
markmap/markmap13.2k
ELI5 (coldteadotai)
Skill把一个代码库、功能、命令或 PR 讲给完全不了解它的人听:先用一句话说清它做什么,再生成 PR Lens 画布,从两个节点开始逐步加入组件,配合分步讲解直到呈现全貌,最后给出可分享的链接。需要同时安装 PR Lens 技能,通过 npx 调用它的 CLI。
coldteadotai/pr-lens2.4k
Excalidraw Diagram Skill
Skill根据文字描述生成手绘风 Excalidraw 图(.excalidraw 文件,可在 excalidraw.com 打开继续编辑),按概念关系安排发散、时间线、汇聚等布局,适合画流程图、概念图和讲解插图。另装 uv 与 Playwright 后 agent 能渲染成图片自查排版,可不装;配色集中在一个文件里,可换成自己的风格。
coleam00/excalidraw-diagram-skill5.2k
Obsidian Visual Skills Pack
Plugin面向 Obsidian 的可视化技能包,把文本生成 Excalidraw 手绘图、Mermaid 图表和 Obsidian Canvas 思维导图,支持中文。作者标注为实验性项目;使用 Excalidraw 技能需装对应的 Obsidian 插件。
axtonliu/axton-obsidian-visual-skills3.7k
ELI5
Skill按听众调整讲法:同一个知识点可以讲给 5 岁孩子、中学生、大学生、家长或管理者,自动换用词、类比、语气和深浅。适合教师分层讲解、家长辅导孩子,也能把专业内容讲给外行听;无需账号或额外软件。
DreambigOu/ELI52k
Nano Banana Gemini CLI Extension
MCPGoogle 官方 Gemini CLI 扩展,内置 MCP 服务器,调用 Nano Banana 模型生成和编辑图片、修复老照片,还能出图标、纹理、分步骤故事板和流程示意图,可用于课件配图与宣传素材。需 Node.js 和 Gemini API Key;在 Claude Code、Codex 中使用要先克隆仓库并构建。
gemini-cli-extensions/nanobanana1.1k