Lighthouse for the AI Agent Era

AgentDoor scans any website and produces an Agent Readiness Score (0–100) measuring how well the site works with AI agents. It identifies specific issues across 5 dimensions and provides actionable fix recommendations.

Just as Google Lighthouse measures website performance for browsers, AgentDoor measures website readiness for AI agents.

AI agents (Claude, GPT, OpenClaw, Gemini) are shifting from answering questions to taking action on the web — placing orders, managing accounts, booking appointments. Websites that work well with agents will capture this traffic. Those that don't will lose it.

AgentDoor answers the simplest question no one has answered yet: "Does my website work well with AI Agents? If not, how do I fix it?"

Grades: A+ (90–100), A (80–89), B (70–79), C (60–69), D (40–59), F (0–39)

• Node.js 18+
• npm

git clone <repo-url> && cd AgentDoor
npm install
npx playwright install chromium

Terminal 1 — Backend API (port 3001):

npm run dev

Terminal 2 — Frontend (port 3000):

cd frontend && npm run dev

Open http://localhost:3000 in your browser.

# Single scan
curl -X POST http://localhost:3001/api/scan \
  -H 'Content-Type: application/json' \
  -d '{"url": "https://stripe.com"}'

# Compare two sites
curl -X POST http://localhost:3001/api/compare \
  -H 'Content-Type: application/json' \
  -d '{"urls": ["https://stripe.com", "https://example.com"]}'

ANTHROPIC_API_KEY=sk-ant-... npm run dev

When set, the scanner uses Claude to generate an executive summary, llms.txt content, and enhanced recommendations. Without it, all 21 rule-based checks still run normally.

Frontend (Next.js :3000)  →  Backend API (Express :3001)  →  Scanner (Playwright)
                                                           →  Claude API (optional)

The backend is stateless — each POST /api/scan triggers the full pipeline:

1. Playwright loads the page (headless Chromium, 15s timeout, images/fonts blocked for speed)
2. Extracts HTML, meta tags, JSON-LD, forms, links, robots.txt, llms.txt, sitemap.xml, MCP endpoint — all in parallel
3. Runs 5 check modules (21 individual checks) to produce dimension scores
4. Scoring engine aggregates into total score + grade
5. (Optional) Claude API generates summary and recommendations

// Request
{ "url": "https://example.com" }

// Response
{
  "url": "https://example.com",
  "scannedAt": "2026-02-24T10:30:00Z",
  "score": 47,
  "grade": "D",
  "label": "Needs Work",
  "dimensions": { ... },
  "topIssues": [ ... ],
  "generatedFixes": { "llmsTxt": "..." },
  "aiSummary": "...",
  "aiRecommendations": [ ... ]
}

// Request
{ "urls": ["https://stripe.com", "https://example.com"] }

// Response
{ "reports": [ <report1>, <report2> ] }

├── src/
│   ├── server.ts                    # Express API server
│   └── scanner/
│       ├── types.ts                 # Shared TypeScript interfaces
│       ├── crawler.ts               # Playwright page loader + data extractor
│       ├── scoring.ts               # Score aggregation + grade mapping
│       ├── ai/analyzer.ts           # Claude API integration (optional)
│       └── checkers/
│           ├── readability.ts       # 5 checks, 25 pts
│           ├── structuredData.ts    # 4 checks, 25 pts
│           ├── navigability.ts      # 4 checks, 20 pts
│           ├── actionability.ts     # 4 checks, 20 pts
│           └── discoverability.ts   # 4 checks, 10 pts
├── frontend/
│   └── app/
│       ├── page.tsx                 # Main UI (scan + compare mode)
│       ├── components/              # ScoreGauge, DimensionBars, IssuesList, etc.
│       └── fixtures/demos.ts        # Pre-scanned demo data for fallback
├── spec.md                          # Full product specification
└── CLAUDE.md                        # AI assistant context

The frontend includes 3 pre-scanned demo fixtures (stripe.com, github.com, example.com) that load instantly. If the backend is unavailable, scanning a demo URL automatically falls back to cached data — useful for presentations without a live internet connection.

MIT