OpenBoard
A local-first whiteboard MCP server that enables AI agents to create, inspect, and update canvas diagrams and shapes collaboratively via 13 semantic tools.
README
<div align="center">
✦ OpenBoard
Local-first personal whiteboard workspace for developers and external AI agents.
<br /> <br />
<p align="center"> <img src="./assets/dashboard.png" alt="OpenBoard Workspace Dashboard" width="100%" /> </p>
<br /> <br />
# Launch directly without installing:
npx openboard-app start
# Or install globally:
npm install -g openboard-app
openboard start
<br />
All whiteboards, documents, and canvas metadata reside locally in your SQLite database:
~/.openboard/openboard.db
<br /> <br />
</div>
✦ Table of Contents
- Overview
- Key Features
- Quick Start
- Model Context Protocol (MCP) Integration
- AI Agent Quick Setup
- 13 Semantic MCP Tools
- Architecture & Guarantees
- Monorepo Structure
- Keyboard Shortcuts
- CLI Reference
- Contributing & Community
- License
✦ Overview
OpenBoard is a modern, lightweight, local-first infinite whiteboard pairing an interactive tldraw canvas with a high-performance Model Context Protocol (MCP) server.
It is designed for developers who want a fast personal whiteboard and for autonomous AI coding agents (Claude Code, Cursor, OpenCode, OpenAI Codex, OpenClaw, Hermes) to inspect, create, organize, and update software architecture diagrams, system topologies, database schemas, and workflows collaboratively.
┌───────────────────────────────────────────────────────────┐
│ EXTERNAL AI AGENT │
│ (Claude Code, Cursor, OpenCode, Codex, Hermes...) │
└─────────────────────────────┬─────────────────────────────┘
│ JSON-RPC 2.0 (stdio / SSE)
▼
┌───────────────────────────────────────────────────────────┐
│ OPENBOARD MCP SERVER │
│ (13 High-Level Semantic Tools) │
└──────────────┬─────────────────────────────┬──────────────┘
│ │
▼ ▼
┌───────────────────────────┐ ┌─────────────────────────────┐
│ LOCAL SQLITE DATABASE │ │ HEADLESS CANVAS ENGINE │
│ (~/.openboard/openboard.db│ │ (tldraw store + SVG vector)│
└───────────────────────────┘ └──────────────┬──────────────┘
│
▼ SSE Live Sync
┌─────────────────────────────┐
│ BROWSER WHITEBOARD │
│ (http://localhost:4747) │
└─────────────────────────────┘
✦ Key Features
- 🔒 Local-First & 100% Private: Zero cloud dependencies, zero telemetry, zero accounts. Everything lives in your local SQLite database (
~/.openboard/openboard.db). - 🤖 AI Agent Native: Built-in stdio & SSE Model Context Protocol server exposing 13 semantic tools for creating and mutating canvas elements.
- 🎨 Twenty-Inspired Dark Workspace: Elegant near-black surfaces (
#0e0e11), electric blue accents (#2563eb), zero-dependency SVG iconography, and responsive design. - 👁️ Headless Canvas Inspection & SVG Screenshots: Agents inspect semantic canvas hierarchies and render pixel-perfect vector SVG snapshots without needing a browser window.
- ⚡ Live Browser Projection (SSE): If a user is viewing a board in their browser, agent modifications stream seamlessly into the viewport in real time.
- ⌨️ Keyboard-First Productivity: Press
Nfor new whiteboards,/to focus search,Escto dismiss dialogs, andEnterto confirm. - 🗃️ Complete Board Lifecycle: Multi-board dashboard, favorites, instant duplication, debounced autosave, soft delete (Trash), and permanent purging.
✦ Quick Start
Option 1: Run with npx (No installation required)
npx openboard-app start
Option 2: Install globally via npm
npm install -g openboard-app
openboard start
Option 3: Connect to your AI Agent immediately
openboard mcp
✦ Model Context Protocol (MCP) Integration
OpenBoard provides a universal Model Context Protocol server over stdio (and HTTP/SSE at http://localhost:4747/api/mcp).
Universal MCP JSON Configuration
Add this to your MCP configuration file:
{
"mcpServers": {
"openboard": {
"command": "openboard",
"args": ["mcp"]
}
}
}
✦ AI Agent Quick Setup
| AI Client | Setup Path or Command | Format |
|---|---|---|
| Claude Code | claude mcp add openboard --command="openboard" --args="mcp" |
CLI / ~/.claude.json |
| Cursor | ~/.cursor/mcp.json |
JSON (mcpServers) |
| OpenCode | ~/.config/opencode/opencode.jsonc |
JSONC (mcp.openboard) |
| OpenAI Codex | ~/.codex/config.toml |
TOML (mcp_servers.openboard) |
| OpenClaw | ~/.openclaw/config.json |
JSON (mcpServers) |
| Hermes Agent | ~/.hermes/config.yaml |
YAML (mcp_servers) |
| Generic MCP | Any client supporting stdio MCP JSON-RPC 2.0 | openboard mcp |
✦ 13 Semantic MCP Tools
OpenBoard equips AI agents with 13 semantic tools structured for developer tasks:
Board Organization & Management
list_boards: Search and filter whiteboards (all,recent,favorites,trash).create_board: Create a new whiteboard with title and optional description.get_board: Retrieve board metadata, document summary, and timestamps.rename_board: Rename an existing whiteboard.duplicate_board: Clone a board and its complete canvas document.favorite_board: Bookmark or toggle favorite status.delete_board: Move a whiteboard to Trash (soft delete).restore_board: Recover a deleted whiteboard from Trash.
Canvas Inspection & Manipulation
get_canvas_state: Retrieve structured shapes, bounds, text, and arrow bindings.get_canvas_screenshot: Render a headless vector SVG visual screenshot.create_shapes: Create shapes (rectangles, ellipses, notes, text, frames, arrows).update_shapes: Modify positions, dimensions, text, colors, or arrow bindings.delete_shapes: Remove shapes and automatically clean up associated connections.
✦ Architecture & Guarantees
- MCP does not connect directly to SQLite: All operations flow through domain-validated
BoardServiceandCanvasService. - MCP does not connect directly to React: Canvas operations execute against a headless tldraw store engine.
- Browser Optional for AI Agents: Agents can create, query, and modify boards whether the web UI is open or closed.
- Explicit
board_idtargeting: Every canvas operation explicitly references the target board ID. - Live Projection: If a user opens the board in a browser, changes made by external agents project into the active view instantly via SSE.
✦ Monorepo Structure
openboard/
├── apps/
│ ├── web/ # React 18 + Vite frontend (tldraw canvas, Twenty dark UI, Docs)
│ └── server/ # Express + SQLite API server + SSE live projection
├── packages/
│ ├── shared/ # TypeScript interfaces, DTOs, schemas & shape contracts
│ ├── storage/ # SQLiteBoardRepository with ACID migrations & memory fallback
│ ├── core/ # BoardService, CanvasService, HeadlessSvgRenderer
│ └── mcp/ # OpenBoardMcpServer (stdio / SSE JSON-RPC 2.0 transport)
├── cli/ # openboard-app CLI binary with embedded web assets
├── docs/ # Architectural guides & AI agent integration tutorials
└── .github/ # CI workflows and issue / PR templates
✦ Keyboard Shortcuts
| Shortcut | Action |
|---|---|
| <kbd>N</kbd> | Open New Whiteboard modal |
| <kbd>/</kbd> | Focus dashboard search bar |
| <kbd>Esc</kbd> | Clear search input / Dismiss modal or context menu |
| <kbd>Enter</kbd> | Submit modal form / Open selected board |
✦ CLI Reference
openboard start # Start local workspace server & open web dashboard (default port: 4747)
openboard mcp # Start Model Context Protocol server on stdio for AI agents
openboard info # Display local configuration, database path, and agent MCP details
openboard -v # Display installed version
openboard --help # Display CLI help
Options for openboard start
-p, --port <number>: Port to listen on (default:4747)-h, --host <host>: Host address to bind to (default:localhost)--db <path>: Custom SQLite database path (default:~/.openboard/openboard.db)--no-open: Do not automatically open browser on startup
✦ Contributing & Community
Contributions are warmly welcome! Whether you are reporting an issue, proposing an MCP feature, or improving developer documentation:
- Read our Contributing Guidelines.
- Check our Code of Conduct.
- Report security disclosures privately per our Security Policy.
# Clone the repository
git clone https://github.com/atpaawej/openboard.git
cd openboard
# Install dependencies & build
npm install
npm run typecheck
npm run test
npm run build
✦ License
OpenBoard is open-source software licensed under the MIT License.
Copyright © 2026 OpenBoard Contributors.
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。