Interoperability layer for AI-enabled websites

Make your website work
with every AI.
From one description.

Describe your site once in a single capability manifest at /ai2w. AI2Web speaks whichever protocol an assistant understands, so agents can read it and act on it safely, without scraping HTML.

Instant AI Readiness score out of 100 · no signup · read the docs →

GET /ai2w

"protocol": "ai2w", "version": "0.2",
"site": { "name": "Northwind", "type": "ecommerce" },
"capabilities": {
  "content": true, "commerce": { "checkout": true },
  "search": true, "events": true, "actions": true
},
"transports": { "mcp", "rest", "acp" },
"auth": { "oauth2": { "pkce": true } },
"consent": { "requires_user_approval_for": ["purchase"] },
"agent_service": { "enabled": true },
"identity": { … }, "contact": { … }, "rate_limits": { … }

✓ Discovery manifest valid✓ MCP transport detected✓ ACP checkout detected✓ Agent-to-Agent (A2A) ready✓ OAuth2 + PKCE · consent gated

100

AI Readiness 100/100Tier · Enterprise

✓

6SDK languages

18open RFCs

/100readiness score

8+agent protocols

What it is

Not "the MCP for websites".
The layer above the protocols.

AI2Web is an open interoperability layer that makes any website understandable and actionable to AI agents. Agents shouldn't scrape HTML and guess at forms: a site describes its capabilities once, then any AI can read and act on them safely, what it is, what it offers, and what actions it supports.

Like a browser that speaks HTTP/1.1, HTTP/2, HTTP/3, WebSockets and WebRTC behind one experience, AI2Web sits above the transports and adds the discovery, validation, governance and analytics that individual protocol specs intentionally leave out.

How it works

One description. Everything downstream.

One capability model

Content, commerce, search, actions, support and more, declared once. Every agent reads the same contract, so you never rebuild per assistant.

Speaks every protocol

The same manifest projects to MCP, ACP, AP2, NLWeb, REST, GraphQL, OpenAPI and feeds. AI2Web negotiates which to use, so any protocol's success is a tailwind.

Discovery + validation

A required /.well-known/ai2w anchor makes sites findable, and the validator scores AI readiness out of 100 with per-capability checks.

Governance + safety

Consent, auth scopes, usage policy and audit are first-class. High-risk actions preview before they run, cross-origin credentials are refused by default.

Transports

One manifest.
Every protocol.

AI2Web sits above the transports and negotiates which to use, so any protocol's success is a tailwind, not a rewrite. Add one and the rest come for free.

source of truth/ai2wYour capability manifest

MCPactions

Tools & actions for assistants

ACPcommerce

Agentic commerce & checkout

AP2payments

Agent payments & signed mandates

NLWebconversational

Natural-language site queries

REST + feedscontent

Content & catalog, cache-friendly

GraphQLquery

Typed queries over capabilities

OpenAPIdescribe

A described, discoverable API

Webhooksevents

Subscribable events, push or poll

Discovery stays neutral too: if a cross-vendor standard emerges, AI2Web is designed to be carried by it rather than compete. WebMCP is supported as an optional in-page adapter.

For assistants

One connector.
Every AI-ready site.

Assistants don't add sites one at a time. Point yours at a single MCP endpoint and it can discover and act on AI-ready sites across the network - no per-site setup, and no new format for anyone to adopt.

Add one endpoint to your assistant

find_sitesdiscover AI-ready sites by capability or type

describe_siteread a site's capabilities and actions

call_site_actioninvoke an action, approval-gated for high-risk

Works with Claude, ChatGPT (developer mode) and any MCP client. The network grows as sites register, so a small, excellent site is reachable on merit - add yours.

Get started