# AI2Web: Open protocol to make any website work with every AI agent
> Source:
> Published: 2026-07-11 00:15:13+00:00
`ai2w`
· one layer for every AI
# Make your website work with every AI - from one place.
AI2Web is the interoperability layer for AI-enabled websites. Describe your website once, and AI2Web automatically exposes its capabilities through the AI protocols your customers use today - MCP, ACP, REST and GraphQL - while staying ready for whatever comes next. Validate, monitor and manage it from one place.
## Don't repeat the Search Console mistake
Website owners already juggle Google Search Console, Bing Webmaster Tools, Yandex, Baidu, social sharing tools and a dozen validators - each with its own portal and rules. Most business owners never learned any of them. The AI era is heading the same way, with MCP, ACP, A2A, discovery manifests, auth profiles and negotiation. It doesn't have to.
**Protocols solve interoperability. AI2Web solves usability, adoption, governance and operations.** It's to AI protocols what Search Console should have been for search: one unified layer instead of a dozen fragmented portals.
## Describe once. Works everywhere.
The web was built for humans. Agents shouldn't have to scrape HTML and guess forms. You declare your capabilities once; AI2Web exposes them through whichever protocols an assistant understands.
### Describe once
One capability manifest for content, commerce, actions, events and auth. No per-assistant rebuilds.
### Generate everything
AI2Web emits MCP, ACP, REST, GraphQL, OpenAPI and feeds from one source of truth. New protocol? Add an adapter, not a rewrite.
### Works today
Backend-first and API-driven. An MCP endpoint works in Claude and ChatGPT connectors right now, with no vendor adoption required.
### Any stack, including plain JavaScript and React
Use AI2Web from TypeScript, plain JavaScript or React, plus PHP, Python, Go, .NET and WordPress. The JavaScript packages ship as standard ESM with bundled types, so they run unchanged in JS. `@ai2web/react`
adds hooks and an embeddable AI Readiness badge, and a build-free ``
web component drops into any HTML page with no bundler.
### Greener by design
Making every AI scrape and render a full web page wastes compute and energy, and AI already consumes a great deal of it. A shared capability layer sends a few kilobytes of structured data instead of megabytes of HTML: less parsing, less bandwidth, less energy. It is the green-internet argument applied to AI, where every request that avoids rendering a page is compute and carbon saved.
## One place to manage your AI presence
Configure your AI capabilities once, and AI2Web keeps them correctly exposed, validated and monitored across the evolving ecosystem - whichever protocols or assistants win.
- ✓
**Enable** AI capabilities in minutes - ✓
**Validate** your implementation and score it - ✓
**Monitor** compatibility across assistants - ✓
**Understand** which capabilities agents actually use - ✓
**Receive** recommendations on what to expose next - ✓
**Manage** everything from one dashboard, not a dozen portals
## Support that just works - the user only has to ask
The highest-value moment is after the sale. A customer asks their assistant; AI2Web handles order tracking, returns and refunds - via MCP, a checkout transport (ACP), or the brand's own agent. Money-moving actions always ask the user to confirm first.
### “Where's my order?”
**→** AI2Web calls `track_order`
(a read, authorised to the user's account) and answers instantly - no confirmation needed.
“DPD - out for delivery, arriving tomorrow ~12:00.”
### “They arrived damaged - refund please.”
**→** `report_issue`
logs it with the photo, then `request_refund`
- a high-risk action - returns a **preview** for the user to approve. Only after approval does it refund, with an audit reference.
“Refund £20.00 to original payment? ✓ Approved · audit aud_01H8”
## The capability model is the product
Everything hangs off one primitive. The framework generates from it, the discovery network indexes by it, the analytics measure against it.
## Complements every protocol - never competes
Like a browser supporting HTTP/1.1, HTTP/2, HTTP/3, WebSockets and WebRTC behind one experience, AI2Web abstracts agent protocols behind one capability model. It defines the model above the transports and negotiates which to use - so any protocol's success is a tailwind.
```
AI Assistant
│
▼
AI2Web
├── MCP
├── ACP
├── REST APIs
├── GraphQL
├── OpenAPI
├── WordPress / CMS abilities
├── Webhooks
└── Future protocols
```
### It's not "the MCP for websites"
AI2Web is the interoperability layer for AI-enabled websites - a common discovery and capability model across protocols, plus the governance, analytics, validation and tooling that protocol specs intentionally leave out. Discovery is pluggable too: if a neutral standard emerges, AI2Web is carried by it rather than competing.
## Frequently asked questions
## What is AI2Web?
AI2Web is an open interoperability layer that makes any website understandable and actionable to AI agents. You describe your site's capabilities once, and AI2Web exposes them across AI assistants and protocols - MCP, ACP, REST, GraphQL and more - from a single place to validate, monitor and manage.
## Does AI2Web replace MCP or ACP?
No - it complements them. Protocols like MCP and ACP solve interoperability; AI2Web sits above them with one common capability model plus the validation, monitoring and management tooling that protocol specifications intentionally leave out. When a protocol wins, AI2Web simply generates it.
## Do I need OpenAI or Anthropic to support AI2Web?
No. The manifest, feeds and endpoints are useful today, and an MCP endpoint already works with existing Claude and ChatGPT connectors. AI2Web ships both the website side and an agent-side connector, so it doesn't depend on any vendor adopting it first.
## Can an AI buy or refund something without my approval?
No. AI2Web is secure by design. Reads such as order tracking happen instantly, but anything that moves money - purchases, refunds, cancellations - is high-risk and always returns a preview for you to approve before it runs, with an audit trail.
## What does AI2Web cost?
The protocol, specification and SDKs are free and open source (MIT for code, CC-BY 4.0 for the spec). Optional hosted endpoints, capability analytics and enterprise features are the commercial layer.
## Which platforms does AI2Web support?
Any website. There's a WordPress and WooCommerce plugin, SDKs for TypeScript, PHP, Python, Go and .NET, React bindings and a framework-free web component for plain JavaScript, and a framework-agnostic server that runs on Node.js or Cloudflare Workers.
## How do I make my website AI-ready?
Add an AI2Web capability manifest at `/ai2w`
using the plugin or an SDK, then validate it to get your AI Readiness score. Agents discover it, negotiate capabilities, and can act - safely, with approval on anything sensitive.
## AI Readiness Validator
Paste an AI2Web manifest (or load an example). Scored exactly like `npx ai2web validate`
- the same logic that ships in the SDKs.
## Help build the AI-ready web
AI2Web is open source and community-driven. The protocol and spec are CC-BY 4.0, the code is MIT. Whether you write code, run a store or maintain a CMS, there is a way in.
### Add an adapter
Map a new protocol or platform into the capability model. The adapter contract is small and enforced by conformance tests.
### Improve an SDK
TypeScript, PHP, Python, Go and .NET all reproduce one conformance contract. Sharpen one, or port to a new language.
### Write docs & examples
Real manifests, framework guides and integration walkthroughs help the next developer ship in minutes.