The open interoperability layer for AI-enabled websites. Describe your site's capabilities once, then expose them across MCP, ACP, REST & GraphQL, validated and managed from one place. MIT-licensed. Works with Claude & ChatGPT connectors today.
No reviews yetBe the first to leave a review for AI2Web
Maker
📌
Hey Product Hunt 👋 I'm Roland.
I was lead dev on a product with 10M+ downloads, and I've spent years watching site owners drown in validators and portals. Lately I've watched every AI vendor ship its own way for agents to "use" a website.
We're about to repeat the Search Console mistake. Remember juggling Search Console, Bing Webmaster Tools, Yandex, and a dozen validators, each with its own portal and its own rules?
The AI web is heading the same way: MCP, ACP, A2A, discovery manifests, auth profiles, negotiation. Most site owners will never learn any of it.
AI2Web is the usability/governance layer the protocols leave out. You describe your site's capabilities once (content, commerce, actions, auth) and it generates MCP, ACP, REST, GraphQL and feeds from that one source of truth.
New protocol?
Add an adapter, not a rewrite.
It complements every protocol and competes with none, so any protocol winning is a tailwind.
What I care about: it works today (an MCP endpoint already runs in Claude & ChatGPT connectors, no vendor adoption needed); it's genuinely open (spec CC-BY 4.0, code MIT, with a WordPress/WooCommerce plugin plus SDKs for TS, PHP, Python, Go, .NET); and it's secure by default (reads like order tracking run instantly, but anything that moves money returns a preview to approve, with an audit trail).
Two things I'd love your honest take on: (1) does "describe once, expose everywhere" hold up against how you're seeing agents actually hit sites, and (2) which adapter should we ship next?
I'm here all day.
Report
💡 Bright idea
How does the validation piece actually work in practice, like is it checking schema compliance on each request or do I need to wire up my own tests against the exposed endpoints?
Report
Maker
@aynurkrbykyt2b Really good question, and it actually made me ship two things, so thank you.
There are three layers:
The validator (design time) fetches your /ai2w manifest and scores it against the spec. It checks the contract you publish, not each live request.
Runtime request validation (new). Every action declares an input_schema, and @ai2web/server now validates each incoming request body against it and returns a 400 before your handler runs. It's on by default, so schema compliance per request is built in, not DIY (validateInput: false to opt out).
Live endpoint checking (new). ai2web check <url> discovers your manifest and probes the declared read endpoints to confirm they respond. It lists your actions but doesn't call them, since those can change state.
So you don't wire up your own schema checks or reachability tests: the validator scores the contract, the server enforces it per request, and check confirms the endpoints are live. You still test your own endpoint logic. Full write-up: ai2web.dev/docs#validation
Report
Maker
Quick update for everyone who supported the AI2Web launch: the AI2Web AI Readiness Checker is now live on the Chrome Web Store.
Click the icon on any site and you will see its AI Readiness Score, tier, and the exact capabilities it is missing for AI agents. It is the fastest way to check a site without leaving the page.
Free, private (only the current tab's origin is sent to score it), no accounts.
How does the validation piece actually work in practice, like is it checking schema compliance on each request or do I need to wire up my own tests against the exposed endpoints?
@aynurkrbykyt2b Really good question, and it actually made me ship two things, so thank you.
There are three layers:
The validator (design time) fetches your /ai2w manifest and scores it against the spec. It checks the contract you publish, not each live request.
Runtime request validation (new). Every action declares an input_schema, and @ai2web/server now validates each incoming request body against it and returns a 400 before your handler runs. It's on by default, so schema compliance per request is built in, not DIY (validateInput: false to opt out).
Live endpoint checking (new). ai2web check <url> discovers your manifest and probes the declared read endpoints to confirm they respond. It lists your actions but doesn't call them, since those can change state.
So you don't wire up your own schema checks or reachability tests: the validator scores the contract, the server enforces it per request, and check confirms the endpoints are live. You still test your own endpoint logic. Full write-up: ai2web.dev/docs#validation
Quick update for everyone who supported the AI2Web launch: the AI2Web AI Readiness Checker is now live on the Chrome Web Store.
Click the icon on any site and you will see its AI Readiness Score, tier, and the exact capabilities it is missing for AI agents. It is the fastest way to check a site without leaving the page.
Free, private (only the current tab's origin is sent to score it), no accounts.
https://chromewebstore.google.com/detail/ai2web-ai-readiness-check/dgfgbjlpmhahoafdcfandobbncfjhlle
Thanks again for the feedback here, a lot of it shaped where this is going.