MindOS

MindOS

Local-first knowledge base MCP server. Lets AI agents (Claude Code, Cursor, etc.) read and write your personal knowledge base through 20 MCP tools. Zero cloud dependency — all files stay on your machine.

Category
Visit Server

README

<p align="center"> <img src="assets/logo-square.svg" alt="MindOS" width="100" /> </p>

<h1 align="center">MindOS</h1>

<p align="center"> <strong>Human Thinks Here, Agent Acts There.</strong> </p>

<p align="center"> <a href="README.md">English</a> | <a href="README_zh.md">中文</a> </p>

<p align="center"> <a href="https://tianfuwang.tech/MindOS"><img src="https://img.shields.io/badge/Website-MindOS-0ea5e9.svg?style=for-the-badge" alt="Website"></a> <a href="https://www.npmjs.com/package/@geminilight/mindos"><img src="https://img.shields.io/npm/v/@geminilight/mindos.svg?style=for-the-badge&color=f59e0b" alt="npm version"></a> <a href="#wechat"><img src="https://img.shields.io/badge/WeChat-Group-07C160.svg?style=for-the-badge&logo=wechat&logoColor=white" alt="WeChat"></a> <a href="LICENSE"><img src="https://img.shields.io/badge/License-MIT-6366f1.svg?style=for-the-badge" alt="MIT License"></a> </p>

MindOS is a Human-AI Collaborative Mind System—a local-first knowledge base that ensures your notes, workflows, and personal context are both human-readable and directly executable by Agents. One shared memory layer for all Agents — auditable, correctable, and smarter with every use.


<p align="center"> <picture> <source media="(prefers-color-scheme: dark)" srcset="assets/images/demo-flow-dark.png" /> <source media="(prefers-color-scheme: light)" srcset="assets/images/demo-flow-light.png" /> <img src="assets/images/demo-flow-light.png" alt="MindOS: From Idea to Execution to Review" width="960" /> </picture> </p>

[!IMPORTANT] ⭐ One-click install: Send this to your Agent (Claude Code, Cursor, etc.) to set up everything automatically:

Help me install MindOS from https://github.com/GeminiLight/MindOS with MCP and Skills. Use English template.

✨ Try it now: After installation, give these a try:

Read my MindOS knowledge base, see what's inside, then help me write my self-introduction into Profile.
Help me distill the experience from this conversation into MindOS as a reusable SOP.
Help me execute the XXX SOP from MindOS.

🧠 Human-AI Shared Mind

No more fragmented memory, no more black-box behavior, no more lost experience.

1. Global Sync — Breaking Memory Silos

Each Agent keeps its own memory — switching tools means manually hauling context. MindOS lets all Agents share one knowledge base via MCP and Skills — record once, reuse everywhere.

2. Transparent & Controllable — No More Black Boxes

What did your Agent remember? Is it even correct? You have no way to know. MindOS saves every read/write as local plain text — humans can audit, correct, and delete in the GUI.

3. Symbiotic Evolution — Experience Flows Back as Instructions

All that experience from your conversations — gone the moment you close the window. MindOS auto-distills conversation experience into Skills/SOPs. Notes are instructions. The knowledge base gets better with use.

Foundation: Local-first by default — all data stays in local plain text for privacy, ownership, and speed.

✨ Features

For Humans

  • GUI Collaboration Workbench: use one command entry to browse, edit, and search efficiently (⌘K / ⌘/).
  • Built-in Agent Assistant: converse in context while edits are captured into managed knowledge.
  • Plugin Views: use scenario-focused views like TODO, Kanban, and Timeline.

For Agents

  • MCP Server + Skills: connect any compatible agent to read, write, search, and run workflows.
  • Structured Templates: start quickly with Profile, Workflows, and Configurations scaffolds.
  • Experience Auto-Distillation: automatically distill daily work into reusable, executable SOP experience.

Infrastructure

  • Security: Bearer Token auth, path sandboxing, INSTRUCTION.md write-protection, atomic writes.
  • Knowledge Graph: visualize relationships and dependencies across notes.
  • Git Time Machine: track every edit, audit history, and roll back safely.
  • Cross-Device Sync: auto-commit, push, and pull via Git — edits on one device appear on all others within minutes.

<details> <summary><strong>Coming Soon</strong></summary>

  • [ ] ACP (Agent Communication Protocol): connect external Agents (e.g., Claude Code, Cursor) and turn the knowledge base into a multi-Agent collaboration hub
  • [ ] Deep RAG integration: retrieval-augmented generation grounded in your knowledge base for more accurate, context-aware AI responses
  • [ ] Backlinks View: display all files that reference the current file, helping you understand how a note fits into the knowledge network
  • [ ] Agent Inspector: render Agent operation logs as a filterable timeline to audit every tool call in detail

</details>


🚀 Getting Started

[!IMPORTANT] Quick Start with Agent: Paste this prompt into any MCP-capable Agent (Claude Code, Cursor, etc.) to install automatically, then skip to Step 3:

Help me install MindOS from https://github.com/GeminiLight/MindOS with MCP and Skills. Use English template.

Already have a knowledge base? Skip to Step 4 to configure MCP + Skills.

1. Install

Option A: npm (recommended)

npm install -g @geminilight/mindos@latest

Option B: Clone from source

git clone https://github.com/GeminiLight/MindOS
cd MindOS
npm install
npm link   # registers the `mindos` command globally

2. Interactive Setup

mindos onboard

The setup wizard will guide you through:

  1. Knowledge base path → default ~/MindOS
  2. Choose template (en / zh / empty / custom)
  3. Ports (Web UI + MCP)
  4. Auth token (auto-generated or passphrase-seeded)
  5. Web UI password (optional)
  6. AI Provider (Anthropic / OpenAI) + API Key — or skip to configure later via mindos config set
  7. Start mode — Background service (recommended, auto-starts on boot) or Foreground

Config is saved to ~/.mindos/config.json automatically.

<!-- For Agents: Before running setup, if the user has not provided LLM configuration, ask them:

  1. AI Provider — anthropic or openai (compatible)?
  2. API Key — the key for the chosen provider
  3. Custom Base URL? — only needed for proxies or OpenAI-compatible endpoints; skip if using official API
  4. Model ID — or use the default Or skip the wizard and edit ~/.mindos/config.json manually (see Config Reference below). -->

<details> <summary>Config Reference (~/.mindos/config.json)</summary>

{
  "mindRoot": "~/MindOS",
  "port": 3000,
  "mcpPort": 8787,
  "authToken": "",
  "webPassword": "",
  "startMode": "daemon",
  "ai": {
    "provider": "anthropic",
    "providers": {
      "anthropic": { "apiKey": "sk-ant-...", "model": "claude-sonnet-4-6" },
      "openai":    { "apiKey": "sk-...",     "model": "gpt-5.4", "baseUrl": "" }
    }
  },
  "sync": {
    "enabled": true,
    "provider": "git",
    "remote": "origin",
    "branch": "main",
    "autoCommitInterval": 30,
    "autoPullInterval": 300
  }
}
Field Default Description
mindRoot ~/MindOS Required. Absolute path to the knowledge base root.
port 3000 Optional. Web app port.
mcpPort 8787 Optional. MCP server port.
authToken Optional. Protects App /api/* and MCP /mcp with bearer token auth. For Agent / MCP clients. Recommended when exposed to a network.
webPassword Optional. Protects the web UI with a login page. For browser access. Independent from authToken.
startMode start Start mode: daemon (background service, auto-starts on boot), start (foreground), or dev.
ai.provider anthropic Active provider: anthropic or openai.
ai.providers.anthropic.apiKey Anthropic API key.
ai.providers.anthropic.model claude-sonnet-4-6 Anthropic model ID.
ai.providers.openai.apiKey OpenAI API key.
ai.providers.openai.model gpt-5.4 OpenAI model ID.
ai.providers.openai.baseUrl Optional. Custom endpoint for proxy or OpenAI-compatible APIs.
sync.enabled false Enable/disable automatic Git sync.
sync.provider git Sync provider (currently only git).
sync.remote origin Git remote name.
sync.branch main Git branch to sync.
sync.autoCommitInterval 30 Seconds after file change to auto-commit+push.
sync.autoPullInterval 300 Seconds between auto-pull from remote.

Multiple providers can be configured simultaneously — switch between them by changing ai.provider. Shell env vars (ANTHROPIC_API_KEY, OPENAI_API_KEY, etc.) take precedence over config file values.

</details>

[!NOTE] If you want the MindOS GUI to be reachable from other devices, make sure the port is open in firewall/security-group settings and bound to an accessible host/network interface.

[!TIP] If you chose "Background service" during onboard, MindOS is installed as a background OS service and starts automatically — no need to run mindos start manually. Run mindos update to upgrade to the latest version.

Open the Web UI in your browser:

mindos open

3. Inject Your Personal Mind with MindOS Agent

  1. Open the built-in MindOS Agent chat panel in the GUI.
  2. Upload your resume or any personal/project material.
  3. Send this prompt: Help me sync this information into my MindOS knowledge base.

<p align="center"> <img src="assets/images/gui-sync-cv.png" alt="Sync CV Example" width="800" /> </p>

4. Make Any Agent Ready (MCP + Skills)

4.1 Configure MindOS MCP

<!-- For Agents: Register MindOS MCP as global (user-level) config by default, so all projects share the same knowledge base. -->

Run one command to auto-install MCP config into your Agent:

mindos mcp install

Supports Claude Code, Cursor, Windsurf, Cline, Trae, Gemini CLI, and more — interactively guides you through agent, scope, transport, and token.

Local (Agent and MindOS on the same machine)

Use stdio transport — no server process needed, most reliable:

# Interactive
mindos mcp install

# One-shot, global scope (shared across all projects)
mindos mcp install -g -y

Remote (Agent on a different machine)

Use http transport — MindOS must be running (mindos start) on the remote machine:

mindos mcp install--transport http --url http://<server-ip>:8787/mcp --token your-token -g

[!NOTE] For remote access, ensure port 8787 is open in your firewall/security-group.

Add -g to install globally — MCP config is shared across all projects instead of the current directory only.

The MCP port defaults to 8787. To use a different port, run mindos onboard and set mcpPort.

<details> <summary>Manual config (JSON snippets)</summary>

Local via stdio (no server process needed):

{
  "mcpServers": {
    "mindos": {
      "type": "stdio",
      "command": "mindos",
      "args": ["mcp"],
      "env": { "MCP_TRANSPORT": "stdio" }
    }
  }
}

Local via URL:

{
  "mcpServers": {
    "mindos": {
      "url": "http://localhost:8787/mcp",
      "headers": { "Authorization": "Bearer your-token" }
    }
  }
}

Remote:

{
  "mcpServers": {
    "mindos": {
      "url": "http://<server-ip>:8787/mcp",
      "headers": { "Authorization": "Bearer your-token" }
    }
  }
}

Each Agent stores config in a different file — see the MCP Config Path column in the Supported Agents table for exact paths.

</details>

4.2 Install MindOS Skills

Skill Description
mindos Knowledge base operation guide (English) — read/write notes, search, manage SOPs, maintain Profiles
mindos-zh Knowledge base operation guide (Chinese) — same capabilities, Chinese interface

Install one skill only (choose based on your preferred language):

# English
npx skills add https://github.com/GeminiLight/MindOS --skill mindos -g -y

# Chinese (optional)
npx skills add https://github.com/GeminiLight/MindOS --skill mindos-zh -g -y

MCP = connection capability, Skills = workflow capability. Enabling both gives the complete MindOS agent experience.

4.3 Common Pitfalls

  • Only MCP, no Skills: tools are callable, but best-practice workflows are missing.
  • Only Skills, no MCP: workflow guidance exists, but the Agent cannot operate your local knowledge base.
  • MIND_ROOT is not an absolute path: MCP tool calls will fail.
  • No authToken set: your API and MCP server are exposed on the network without protection.
  • No webPassword set: anyone who can reach your server can access the web UI.

⚙️ How It Works

A fleeting idea becomes shared intelligence through three interlocking loops:

graph LR
    H["👤 Human<br/><sub>thinks · reviews · evolves</sub>"]
    M[("📚 MindOS")]
    A["🤖 Agent<br/><sub>executes · retrospects · extracts SOPs</sub>"]
    EXT["🌐 All Agents"]

    H -- "ideas & feedback" --> M
    M -- "context & insights" --> H
    M -- "instructions & context" --> A
    A -- "results & SOPs" --> M
    M -. "via MCP" .-> EXT

    style H fill:#f59e0b,stroke:#d97706,color:#fff,stroke-width:2px
    style M fill:#10b981,stroke:#059669,color:#fff,stroke-width:2px
    style A fill:#6366f1,stroke:#4f46e5,color:#fff,stroke-width:2px
    style EXT fill:#64748b,stroke:#475569,color:#fff,stroke-dasharray:5 5

Both sides evolve. Humans gain new insights from accumulated knowledge; Agents extract SOPs and get smarter. MindOS sits at the center — the shared second brain that grows with every interaction.

Collaboration Loop (Human + Multi-Agent)

  1. Human reviews and updates notes/SOPs in the MindOS GUI (single source of truth).
  2. Other Agent clients (OpenClaw, Claude Code, Cursor, etc.) connect through MCP and read the same memory/context.
  3. With Skills enabled, those Agents execute workflows and SOP tasks in a guided way.
  4. Execution results are written back to MindOS so humans can audit and refine continuously.

Who is this for?

  • Independent Developer — Store personal SOPs, tech stack preferences, and project context in MindOS. Any Agent instantly inherits your work habits.
  • Knowledge Worker — Manage research materials with bi-directional links. Your AI assistant answers questions grounded in your full context, not generic knowledge.
  • Team Collaboration — Share a MindOS knowledge base across team members as a single source of truth. Humans and Agents read from the same playbook, keeping everyone aligned.
  • Automated Agent Operations — Write standard workflows as Agent-Ready documents. Agents execute directly, humans audit the results.

🤝 Supported Agents

Agent MCP Skills MCP Config Path
MindOS Agent Built-in (no config needed)
OpenClaw ~/.openclaw/openclaw.json or ~/.openclaw/mcp.json
Claude Desktop macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Claude Code ~/.claude.json (global) or .mcp.json (project)
CodeBuddy ~/.claude-internal/.claude.json (global)
Cursor ~/.cursor/mcp.json (global) or .cursor/mcp.json (project)
Windsurf ~/.codeium/windsurf/mcp_config.json
Cline macOS: ~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json; Linux: ~/.config/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json
Trae ~/.trae/mcp.json (global) or .trae/mcp.json (project)
Gemini CLI ~/.gemini/settings.json (global) or .gemini/settings.json (project)
GitHub Copilot .vscode/mcp.json (project) or VS Code User settings.json (global)
iFlow iFlow platform MCP configuration panel

📁 Project Structure

MindOS/
├── app/              # Next.js 16 Frontend — Browse, edit, and interact with AI
├── mcp/              # MCP Server — HTTP adapter that maps tools to App API
├── skills/           # MindOS Skills (`mindos`, `mindos-zh`) — Workflow guides for Agents
├── templates/        # Preset templates (`en/`, `zh/`, `empty/`) — copied to knowledge base on onboard
├── bin/              # CLI entry point (`mindos onboard`, `mindos start`, `mindos open`, `mindos sync`, `mindos token`)
├── scripts/          # Setup wizard and helper scripts
└── README.md

~/.mindos/            # User data directory (outside project, never committed)
├── config.json       # All configuration (AI keys, port, auth token, sync settings)
├── sync-state.json   # Sync state (last sync time, conflicts)
└── my-mind/          # Your private knowledge base (default path, customizable on onboard)

⌨️ CLI Commands

Command Description
mindos onboard Interactive setup (config, template, start mode)
mindos start Start app + MCP server (foreground, production mode)
mindos start --daemon Install + start as a background OS service (survives terminal close, auto-restarts on crash)
mindos dev Start app + MCP server (dev mode, hot reload)
mindos dev --turbopack Dev mode with Turbopack (faster HMR)
mindos open Open the Web UI in the default browser
mindos stop Stop running MindOS processes
mindos restart Stop then start again
mindos build Manually build for production
mindos mcp Start MCP server only
mindos token Show auth token and per-agent MCP config snippets
mindos sync Show sync status (alias for sync status)
mindos sync init Interactive setup for Git remote sync
mindos sync status Show sync status: last sync, unpushed commits, conflicts
mindos sync now Manually trigger a full sync (commit + push + pull)
mindos sync on Enable automatic sync
mindos sync off Disable automatic sync
mindos sync conflicts List unresolved conflict files
mindos gateway install Install background service (systemd on Linux, LaunchAgent on macOS)
mindos gateway uninstall Remove background service
mindos gateway start Start the background service
mindos gateway stop Stop the background service
mindos gateway status Show background service status
mindos gateway logs Tail background service logs
mindos doctor Health check (config, ports, build, daemon status)
mindos update Update MindOS to the latest version
mindos logs Tail service logs (~/.mindos/mindos.log)
mindos config show Print current config (API keys masked)
mindos config validate Validate config file
mindos config set <key> <val> Update a single config field
mindos Start using the mode saved in ~/.mindos/config.json

⌨️ Keyboard Shortcuts

Shortcut Function
⌘ + K Global Search
⌘ + / Call AI Assistant / Sidebar
E Press E in View mode to quickly enter Edit mode
⌘ + S Save current edit
Esc Cancel edit / Close dialog

💬 Community <a name="wechat"></a>

Join our WeChat group for early access, feedback, and AI workflow discussions:

<p align="center"> <img src="assets/images/wechat-qr.png" alt="WeChat Group QR Code" width="200" /> </p>

Scan the QR code or ask an existing member to invite you.


📄 License

MIT © GeminiLight

Recommended Servers

playwright-mcp

playwright-mcp

A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.

Official
Featured
TypeScript
Magic Component Platform (MCP)

Magic Component Platform (MCP)

An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.

Official
Featured
Local
TypeScript
Audiense Insights MCP Server

Audiense Insights MCP Server

Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.

Official
Featured
Local
TypeScript
VeyraX MCP

VeyraX MCP

Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.

Official
Featured
Local
graphlit-mcp-server

graphlit-mcp-server

The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.

Official
Featured
TypeScript
Kagi MCP Server

Kagi MCP Server

An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.

Official
Featured
Python
E2B

E2B

Using MCP to run code via e2b.

Official
Featured
Neon Database

Neon Database

MCP server for interacting with Neon Management API and databases

Official
Featured
Exa Search

Exa Search

A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.

Official
Featured
Qdrant Server

Qdrant Server

This repository is an example of how to create a MCP server for Qdrant, a vector search engine.

Official
Featured