Nabab

What it does

Nabab renders Bayesian networks as interactive D3-based graphs where you can drag probability sliders, toggle hard and soft evidence on nodes, inspect conditional probability tables, and instantly see updated posterior distributions — all in the browser.

Key features

• Multiple inference algorithms — Junction Tree (exact), Variable Elimination, Loopy Belief Propagation (approximate), and a cached engine that reuses junction tree structure for up to 8× faster repeated queries
• Interactive D3 viewer — Dagre auto-layout, probability bar sliders, hard/soft evidence toggling, CPT inspection panel, drag-and-drop XMLBIF loading, and URL state persistence
• Worker-based inference — Runs in a Web Worker (browser) or worker_threads (Node.js) to keep the UI responsive on large networks
• 17 built-in benchmark models — Asia, Alarm, Sachs, Child, Insurance, Hailfinder, Andes, and more from bnlearn, all loadable from the toolbar
• MCP server — Lets Claude and other LLMs load networks, set evidence, and query posteriors via tool calls

Installation

Claude Desktop: Add to claude_desktop_config.json:

{
  "mcpServers": {
    "nabab": {
      "command": "npx",
      "args": ["tsx", "src/mcp/server.ts"],
      "cwd": "/path/to/nabab"
    }
  }
}

Browser viewer:

git clone https://github.com/ochafik/nabab.git && cd nabab
npm install && npm run dev

Supported hosts

Claude Desktop