mindkeeper-mcp
Captures ideas from conversations and organizes them into a persistent, hierarchical mindmap. Supports search, deduplication, export, import, and cloud sync.
README
mindkeeper-mcp
An MCP server that captures ideas from conversations and organises them into a persistent mindmap. Ideas are stored as nodes that can be linked to a parent, tagged, searched, exported, and synced to the cloud — surviving across sessions.
Data is stored in ~/.mindkeeper/mindmap.json with atomic writes and automatic backups.
Features
- Persistent — mindmap survives across conversations and restarts
- Hierarchical — nest ideas under parents to build tree structure
- Searchable — weighted full-text search across text and tags
- Deduplication — same idea under the same parent is never added twice
- Safe writes — atomic temp-file → backup → rename strategy
- Concurrency-safe — serialised write queue prevents file corruption
- Export — Markdown, Mermaid diagram, OPML, JSON, or interactive HTML (with PNG/SVG download)
- Import — build a mindmap from your Claude.ai conversation history
- Cloud sync — backup and restore via private GitHub Gist
Installation
Global install (recommended)
npm install -g mindkeeper-mcp
Local install
git clone https://github.com/icedsg/mindkeeper-mcp
cd mindkeeper-mcp
npm install
npm run build
Configuration
Claude Desktop
Edit claude_desktop_config.json. The quickest way to open it: in Claude Desktop, go to Settings → Developer → Edit Config.
Alternatively, find the file at:
| Platform | Location |
|---|---|
| macOS | ~/Library/Application Support/Claude/claude_desktop_config.json |
| Windows | %APPDATA%\Claude\claude_desktop_config.json |
| Linux | ~/.config/claude/claude_desktop_config.json |
Global install:
{
"mcpServers": {
"mindkeeper": {
"command": "mindkeeper-mcp"
}
}
}
Local install (absolute path):
{
"mcpServers": {
"mindkeeper": {
"command": "node",
"args": ["/absolute/path/to/mindkeeper-mcp/build/index.js"]
}
}
}
After editing the config, restart Claude Desktop for the server to connect.
Claude Code (CLI)
# Global install
claude mcp add mindkeeper -- mindkeeper-mcp
# Local install
claude mcp add mindkeeper -- node /path/to/mindkeeper-mcp/build/index.js
Automatic topic capture (recommended)
Add this to your Claude Desktop system prompt (Settings → Profile → Custom Instructions) to make mindkeeper capture your topics automatically:
You have mindkeeper-mcp connected.
- After each of my messages, if I mention a new topic, question, goal, or interest — call add_idea to record it. Only capture what I say, never your own responses.
- Before adding, call search_ideas to avoid duplicates.
Tools
| Tool | Description |
|---|---|
add_idea |
Capture a new idea, optionally attached to a parent node |
update_node |
Edit the text or tags of an existing idea |
delete_node |
Remove an idea; children are orphaned (kept, not deleted) |
search_ideas |
Full-text search across idea text and tags |
get_mindmap |
Retrieve the full tree, or a subtree from a given node |
export_markdown |
Export as a nested Markdown list |
export_mermaid |
Export as a Mermaid flowchart — paste into GitHub, Notion, or Obsidian |
export_opml |
Export as OPML — import into MindNode, OmniOutliner, or XMind |
export_json |
Export raw JSON — use with the online visualizer |
export_html |
Generate a self-contained interactive HTML file saved to ~/.mindkeeper/mindmap-export.html — open in browser, drag, zoom, export PNG or SVG |
import_claude_export |
Parse a conversations.json from Claude.ai's data export and build a mindmap from your conversation history |
sync_cloud |
Push or pull the mindmap to/from a private GitHub Gist |
cloud_status |
Show current cloud sync configuration |
Usage examples
Capture a top-level idea:
add_idea text="Product strategy for Q3" tags=["strategy","q3"]
Add a sub-idea under an existing node:
add_idea text="Launch in EU market" parentId="<id from previous call>" tags=["launch"]
Refine an idea:
update_node nodeId="<id>" newText="Launch in EU market — target Germany first"
Find related ideas:
search_ideas query="EU launch"
See the whole map:
get_mindmap
Export for a document:
export_markdown
Visualize interactively:
export_html # opens in browser — drag, zoom, download PNG or SVG from the toolbar
Or paste into an online renderer:
export_json # drop the output into the visualizer at the project website
Cloud sync
Create ~/.mindkeeper/config.json:
{
"cloud": {
"provider": "github_gist",
"token": "ghp_YOUR_PERSONAL_ACCESS_TOKEN"
}
}
Generate a token at github.com/settings/tokens with the gist scope. The first sync_cloud direction="push" auto-creates a private Gist and saves the gistId back to config.
Visualizer
Export your mindmap as JSON and drop it into the browser-based visualizer at icedsg.github.io/mindkeeper-mcp/visualize.html for zoomable, draggable, multi-layout exploration.
Import from Claude.ai
Turn your entire Claude.ai conversation history into a structured mindmap in three steps.
Step 1 — Export your Claude.ai data
- Open claude.ai and go to Settings → Account
- Scroll to Export Data and click Export
- Claude emails you a download link within a few minutes
- Download the ZIP and unzip it — find
conversations.jsoninside
Step 2 — Import and build the mindmap
Tell Claude:
Import my Claude export from /path/to/conversations.json into my mindmap
Replace the path with the actual location:
- Windows:
C:\Users\YourName\Downloads\claude-export\conversations.json - Mac/Linux:
/Users/YourName/Downloads/claude-export/conversations.json
Claude reads every conversation, clusters them by theme, and populates the mindmap. Trivial or very short conversations are skipped automatically.
Step 3 — Visualise the result
Export the mindmap as HTML
The HTML file is saved to ~/.mindkeeper/mindmap-export.html and opens in any browser — no server needed. Inside the file, use the toolbar buttons to download a PNG or SVG image.
Data storage
The mindmap is stored in ~/.mindkeeper/mindmap.json. A backup is kept at ~/.mindkeeper/mindmap.json.bak and is overwritten on every save.
To reset: delete or rename mindmap.json. The server creates a fresh empty map on next use.
Development
npm run dev # tsx watch mode — restarts on file change
npm run build # compile TypeScript to ./build
npm test # run test suite (requires bash)
Troubleshooting
Server not found after global install
Ensure npm's global bin directory is on your PATH:
npm config get prefix # e.g. /usr/local
# Add /usr/local/bin to PATH if missing
Permission denied on ~/.mindkeeper
mkdir -p ~/.mindkeeper
chmod 755 ~/.mindkeeper
Mindmap is empty after restart
Check the file exists and is valid JSON:
cat ~/.mindkeeper/mindmap.json | node -e "process.stdin.resume(); let d=''; process.stdin.on('data',c=>d+=c); process.stdin.on('end',()=>console.log(JSON.parse(d).rootId))"
If the file is corrupt, restore from backup:
cp ~/.mindkeeper/mindmap.json.bak ~/.mindkeeper/mindmap.json
stdio errors in Claude Desktop logs
The server logs all operations to stderr (visible in Claude Desktop's MCP logs). Normal log lines start with [mindkeeper]. Anything else is an unexpected error.
License & Credits
MIT License — see LICENSE.
Open-source dependencies:
| Package | License | Used for |
|---|---|---|
| @modelcontextprotocol/sdk | MIT | MCP server protocol |
| uuid | MIT | Node ID generation |
| D3.js | ISC | Force graph & tree views in the web visualizer |
Original work:
mindkeeper-map.js — the interactive mindmap renderer used in the exported HTML and web visualizer is original code with no external dependencies. Algorithms: slot-based tree layout, cubic-bezier links, canvas measureText node sizing, per-node drag coexisting with canvas pan.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。