aiendpoint.dev
Open Standard · IETF Internet-Draft · Apache 2.0

The /ai Standard

The web was built for human browsers.
AI agents are a fundamentally different client.

Every time an AI agent visits your site, it processes 10,000–50,000 tokens — HTML, scripts, styles, ads — just to find the few hundred tokens it actually needed./ai ends that.

robots.txt· don't go
sitemap.xml· go here
llms.txt· context
/ai· do this
Why we built this →
AI agent reads a webpage
18% HTML structure + content
42% JavaScript bundles
20% CSS & styles
12% Ads, trackers, widgets
5% Useful information

10k–50ktokens

varies by page complexity · mostly noise

AI agent reads /ai
20% Service name & description
55% Capabilities & endpoints
15% Auth & rate limits
10% Token hints

~800tokens

pure signal · 0% noise

60× fewer tokens. Zero parsing. Instant, reliable access to exactly what the agent needs. Full story →

0

Indexed services

1.0

Spec version

10min

To implement

Works on every website

Install the MCP server once. Your AI agent auto-discovers what any site can do — even if the site hasn't implemented /ai yet.

1

Direct check

Agent fetches site.com/ai directly. If the site implements the standard, done — ~800 tokens, 100% accurate.

2

Community registry

No /ai? Check if another agent already generated a spec. 100+ popular services are pre-indexed and growing.

3

Agent generates

Still nothing? Your agent analyzes the site and creates the spec. It's saved to the registry — every future agent benefits.

Your AI agent pays the generation cost once. The community shares the result forever. Learn more →

Start here

Choose what fits you best

Minimal /ai response

GET /ai → 200 OK · application/jsonValidate your /ai →
{
  "aiendpoint": "1.0",
  "service": {
    "name": "My Service",
    "description": "Concise description for AI agents",
    "category": ["productivity"]
  },
  "capabilities": [
    {
      "id": "search_items",
      "description": "Search items by keyword",
      "endpoint": "/api/search",
      "method": "GET",
      "params": { "q": "keyword (string, required)" },
      "returns": "items[] with id, name, price"
    }
  ]
}

Use with your AI agent

Auto-discover any website's capabilities from Claude, Cursor, or Claude Code.

Full setup guide →

Claude Desktop

{
  "mcpServers": {
    "aiendpoint": {
      "command": "npx",
      "args": ["-y",
        "@aiendpoint/mcp-server"]
    }
  }
}

Cursor

{
  "mcpServers": {
    "aiendpoint": {
      "command": "npx",
      "args": ["-y",
        "@aiendpoint/mcp-server"]
    }
  }
}

Claude Code

skills.sh ↗
npx skills add aiendpoint/platform --skill aiendpoint

Adds /ai to your own service from inside Claude Code.

Open standard · Apache 2.0

The web got better when everyone added robots.txt.
It's happening again.

Every service that implements /ai reduces wasted tokens across the entire AI ecosystem — not for one company, but for everyone building with AI. The spec is open. Implementation takes 10 minutes.

Why this matters →