io.github.HomenShum/nodebench

Name: io.github.HomenShum/nodebench
Rating: 0.7 (13 reviews)
Author: homenshum

编码与调试

by homenshum

覆盖 49 个领域的 260 个 MCP 工具，支持 AI Flywheel、质量门禁、研究与 web scraping。

13GitHub

什么是 io.github.HomenShum/nodebench？

覆盖 49 个领域的 260 个 MCP 工具，支持 AI Flywheel、质量门禁、研究与 web scraping。

README

NodeBench AI

Entity intelligence for any company, market, or question.

Live: nodebenchai.com · npm: npx nodebench-mcp · GitHub: HomenShum/nodebench-ai

What It Does

Search any company. Get a decision-ready intelligence packet with people, timeline, financials, competitive landscape, product intelligence, and risk flags — shaped for your role (founder, investor, banker, CEO, legal, student).

Deep diligence: 6 parallel research branches, each chaining up to 3 levels deep
Gap remediation: Every risk comes with actionable steps, effort estimates, and expected outcomes
SEO audit: Automatic discoverability scoring with missing-presence detection
Self-search: Search your own company — NodeBench injects your local context for honest self-assessment

Quick Start

Option 1: Use the Web App

Go to nodebenchai.com and search.

Option 2: Connect via MCP (Claude Code / Cursor / Windsurf)

bash

# Claude Code (one command)
claude mcp add nodebench -- npx -y nodebench-mcp --preset founder

# Cursor
npx nodebench-mcp --preset cursor

# Any MCP client
npx nodebench-mcp --preset starter

That's it. NodeBench starts with 15 tools and discovers more as you need them.

Option 2b: Claude Code Plugin (slash commands + Codex delegation)

bash

/plugin marketplace add HomenShum/nodebench-ai
/plugin install nodebench@nodebench
/reload-plugins
/nodebench:setup

Commands: /nodebench:search, /nodebench:diligence, /nodebench:remediate, /nodebench:packet

With Codex plugin installed, /nodebench:remediate --delegate sends gap fixes to Codex for background implementation.

Option 3: Run Everything Locally

bash

git clone https://github.com/HomenShum/nodebench-ai.git
cd nodebench-ai
npm install
cp .env.example .env.local  # Add your API keys

# Start the app
npm run dev                  # Frontend (Vite, port 5191)
npx convex dev               # Backend (Convex)

# Or just the MCP server
cd packages/mcp-local && npx tsx src/index.ts

Claude Code Setup Guide

After running claude mcp add nodebench -- npx -y nodebench-mcp --preset founder, Claude Code can guide itself. Here's what to tell it:

code

I have NodeBench MCP connected. Help me:

1. Run `discover_tools` to see what's available
2. Search my company: use `web_search` + `enrich_entity` for "[Your Company]"
3. Get my weekly reset: use `founder_local_weekly_reset`
4. Analyze a competitor: use `run_recon` for "[Competitor Name]"

Presets

Preset	Tools	Best for
`starter`	15	First-time users, any IDE
`founder`	~40	Founders — weekly reset, delegation, company truth
`banker`	~40	Bankers — diligence, credit memo, risk analysis
`cursor`	28	Cursor IDE (fits tool cap)
`full`	350+	Power users — everything

Architecture

code

nodebenchai.com (React + Vite + Tailwind)
    ↓
Convex Cloud (realtime DB + 10-min actions + durable workflows)
    ↓
Deep Diligence Pipeline:
  Entity Resolution → 6 Parallel Branches → Chained Depth (3 levels)
    ├── People & Leadership
    ├── Company History & Timeline
    ├── Financials & Metrics
    ├── Market & Competitive
    ├── Products & Technology
    └── Risks & Diligence Flags
    ↓
Gap Remediation → SEO Audit → Actionable Next Steps
    ↓
Result Packet (realtime via Convex subscription)

Key Tech

Frontend: React, Vite, TypeScript, Tailwind CSS
Backend: Convex (realtime database, serverless functions, durable workflows)
Search: Linkup API + Gemini 3.1 extraction + 4-layer grounding pipeline
MCP Server: Node.js, TypeScript, better-sqlite3, 350+ tools across 57 domains
Design: Glass cards, terracotta #d97757, Manrope + JetBrains Mono

API Keys

Set these in .env.local (local dev) or Convex environment (production):

Key	Required	Purpose
`GEMINI_API_KEY`	Yes	Gemini 3.1 for classification, extraction, synthesis
`LINKUP_API_KEY`	Recommended	Deep web search with sourced answers
`VITE_CONVEX_URL`	Yes (app)	Convex deployment URL

bash

# Set Convex env vars
npx convex env set GEMINI_API_KEY "your-key"
npx convex env set LINKUP_API_KEY "your-key"

Project Structure

code

nodebench-ai/
├── src/                          # React frontend
│   ├── features/                 # Feature modules (controlPlane, founder, monitoring)
│   ├── hooks/useConvexSearch.ts  # Convex search hook (realtime polling)
│   └── layouts/                  # App shell, surface routing
├── convex/                       # Convex backend
│   ├── domains/search/           # Deep diligence pipeline
│   │   ├── searchPipeline.ts     # Mutations + queries (start, get, cache)
│   │   ├── searchPipelineNode.ts # Quick search action
│   │   └── deepDiligence.ts      # 6-branch deep diligence + remediation
│   └── schema.ts                 # Database schema (50+ tables)
├── packages/mcp-local/           # MCP server (npm: nodebench-mcp)
│   ├── src/tools/                # 350+ tool implementations
│   ├── src/subconscious/         # Memory blocks, graph engine, whisper policy
│   └── src/toolsetRegistry.ts    # Lazy-loading tool domains
├── server/                       # Express server (local dev + Vercel)
│   ├── routes/search.ts          # SSE search (Vercel fallback)
│   └── agentHarness.ts           # Agent orchestration
└── docs/architecture/            # Specs and analysis

License

MIT

常见问题