Hey Product Hunt! I built AgentWeb because I kept hitting the same wall: AI agents have no way to discover what a website can do. There's no robots.txt equivalent that says "here are my capabilities, here's how to authenticate, here's when to escalate to a human." I scored 100 sites and the average was 40/100. Every single one failed. Three things you can try right now: 1. Score your site: `npx agentweb score https://yoursite.com` 2. Generate a starter agent.json: `npx agentweb init --industry retail` 3. Proxy any site as an MCP server: `npx @agentweb-dev/middleware --origin https://yoursite.com` The middleware is the one I'm most excited about. Point it at any website and it creates an MCP server in front of it with tools for browsing, search, structured data extraction, and human escalation. No code changes needed. Everything is open source and on npm. Would love feedback on the scoring methodology and the spec itself.

AgentWeb

Score any website's AI agent readiness. Fix it in minutes.

Score any website's AI agent readiness. Fix it in minutes.