Wondering how to use WebMCP? The quick version is just below — and the timing matters, because on 25 August 2026 OpenAI switched on WebMCP site tools in the ChatGPT desktop app’s built-in browser, which means websites that expose tools are suddenly first-class citizens for ChatGPT Work and Codex, while everyone else’s site is still being screen-scraped.
Short answer:
- WebMCP is an experimental open standard that lets your website register JavaScript functions as structured tools AI agents can call directly, instead of guessing at your buttons from the HTML.
- OpenAI announced ChatGPT support on 25 August 2026: ChatGPT Work and Codex discover and use site tools in the desktop app’s built-in browser.
- You build tools two ways in Chrome: an imperative API (register standard JavaScript functions) or a declarative API (annotate existing HTML forms).
- It’s live in Chrome 149+ via origin trial; test locally with the chrome://flags/#enable-webmcp-testing flag.
- OpenAI is running a 10-day WebMCP Challenge ($35,000 in prizes, partners include Google Chrome, Cloudflare, Shopify, Vercel and Render) — submissions close 3 September 2026.
Julian covered the ChatGPT Work update that shipped WebMCP site tools in this video:
What WebMCP Actually Is (and Why OpenAI Just Made It Matter)
Two primary sources for everything in this guide: OpenAI’s announcement on its developer forum, published 25 August 2026 (Build agent-ready websites with ChatGPT), and Chrome’s developer documentation for WebMCP (developer.chrome.com/docs/ai/webmcp).
Until now, an AI agent using your website had to read your HTML, guess which button does what, and simulate clicks — slow, brittle, and wrong often enough to be scary. WebMCP flips the relationship: your page registers named tools with descriptions and JSON Schema input definitions, and the agent calls them like an API. OpenAI calls it “an experimental open standard that lets web apps expose tools that agents can use directly on the page.”
The OpenAI side of this: in the built-in browser of the ChatGPT desktop app, ChatGPT Work and Codex can discover and use these tools when they’re available. No extension, no integration deal — if your site speaks WebMCP, their agents can use it.
How to Use WebMCP on Your Own Website
Chrome’s developer guide gives you two routes in:
| Approach | How it works | Best for |
|---|---|---|
| Imperative API | Register standard JavaScript functions as tools — form input, navigation, state management, or any function you choose — each with a name, description and JSON Schema inputs | Web apps with real logic: checkouts, dashboards, booking flows |
| Declarative API | Add annotations to standard HTML forms so the form itself becomes a WebMCP tool | Content sites, contact and lead forms, anything mostly static |
The practical first steps, straight from the Chrome docs:
- Get a build of Chrome that supports it. WebMCP is in Chrome 149+ through an origin trial; for local development, open
chrome://flags/#enable-webmcp-testingand enable the flag. - Pick your tools deliberately. Start with one or two high-value actions — a search, a booking, a quote form. Explicit input and output definitions via JSON Schema exist, in Chrome’s words, “to reduce hallucination or misunderstanding”.
- Test with the inspector. Google ships a Model Context Tool Inspector extension so you can see and call the tools your page registers before any agent does.
- Steal from the demos. Google publishes three: Pizza Maker and a React travel demo (imperative API), and Le Petit Bistro (declarative API).
🔥 Want this set up without the guesswork? Making your site agent-ready with WebMCP is exactly the kind of play we implement together inside the AI Profit Boardroom — 3,700+ members, four live calls a week, daily tutorials, done-for-you templates and a 30-day roadmap. Want a personal look at whether your site should expose tools yet? Book a free SEO strategy session and we’ll go through it live.
How to Use WebMCP Safely: The Security Gates
Handing agents callable functions on your site sounds terrifying, so it’s worth knowing WebMCP ships with two gatekeepers baked in (per the Chrome documentation):
- Origin isolation. WebMCP is only available in origin-isolated documents — a page served with the
Origin-Agent-Cluster: ?0header disables access entirely. - Permissions Policy. The whole feature is gated by the
toolspolicy, which defaults toself(same-origin only). A cross-origin iframe can’t touch your tools unless you explicitly grant itallow="tools".
And the obvious one: agents can only call functions you chose to register. A read-only search tool and a “fill this enquiry form” tool are very different risk levels to a “delete account” tool — start read-only, add write actions once you trust the flow.
Why SEO People Should Care Right Now
This is the part I care about most. For twenty years “being findable” meant being crawlable by Google. WebMCP is the same shift for the agent era: a site with registered tools is one an agent can actually transact with, while a competitor’s site is a wall of HTML the agent has to guess about. The OpenAI announcement’s own framing — “build agent-ready websites” — tells you where this is going, and the partner list for the challenge (Google Chrome, Cloudflare, Shopify, Vercel, Render, Netlify) tells you the infrastructure players agree.
There’s also a deadline if you want to learn by shipping: OpenAI’s 10-day WebMCP Challenge, announced 25 August 2026, offers $35,000 in cash prizes plus Codex Micros and ChatGPT Pro subscriptions, and submissions close 3 September 2026.
If agents driving websites is new territory for you, these fit well next: my breakdown of Yutori Navigator n2 (a computer-use model that browses the old way), my guide to AutoClaw, and my look at the Magnitude AI agent.
One honest limit: the OpenAI announcement itself is light on implementation detail — it points at the WebMCP specification and Chrome’s guide rather than shipping its own docs — so Chrome’s documentation is where the real build instructions live today. I’d expect richer OpenAI-side docs to follow; when they land, this guide gets updated.
The bottom line on how to use WebMCP
Register a couple of genuinely useful functions as tools (imperative API for app logic, declarative annotations for forms), test them in Chrome 149+ with the testing flag and the Model Context Tool Inspector, keep the security defaults on, and your site becomes something ChatGPT Work and Codex can operate rather than scrape. It’s experimental, the docs are young, and none of this replaces normal SEO — but the sites that expose clean tools early will be the ones agents prefer to transact with, and that’s a first-mover window I wouldn’t sleep on.
FAQ: how to use WebMCP
What is WebMCP?
WebMCP is an experimental open standard that lets a website expose JavaScript functions as structured tools that AI agents can call directly on the page, instead of guessing at buttons from the HTML. OpenAI added support to the ChatGPT desktop app’s built-in browser on 25 August 2026.
Which ChatGPT products support WebMCP site tools?
Per OpenAI’s announcement, ChatGPT Work and Codex can discover and use WebMCP site tools inside the built-in browser of the ChatGPT desktop app when a site exposes them.
Do I need Chrome to build with WebMCP?
Chrome is the main development target right now: WebMCP is available in Chrome 149+ through an origin trial, and locally by enabling the chrome://flags/#enable-webmcp-testing flag. The standard itself is an open proposal, not a Chrome-only feature.
How do I test WebMCP tools locally?
Enable the WebMCP testing flag in Chrome, then use the Model Context Tool Inspector extension to see and call the tools your page registers. Google also publishes demo apps — Pizza Maker and a React travel demo for the imperative API, and Le Petit Bistro for the declarative API.
Is WebMCP secure?
It ships with two gatekeepers: WebMCP only works in origin-isolated documents, and it is gated by the ‘tools’ Permissions Policy, which defaults to same-origin — a cross-origin iframe needs an explicit allow=”tools” attribute. You choose exactly which functions to expose.
What is the OpenAI WebMCP Challenge?
A 10-day hackathon OpenAI announced on 25 August 2026, with $35,000 in cash prizes plus Codex Micros and ChatGPT Pro subscriptions, run with partners including Google Chrome, Cloudflare, Shopify, Vercel, Render and Netlify. Submissions close 3 September 2026.
Related reading
- Yutori Navigator n2: 27B Computer-Use Model Explained
- What Is AutoClaw? Z.ai’s Free AI Agent App Explained
- Magnitude AI Agent: Full Guide
Next steps: if you want your site agent-ready before your competitors even know WebMCP exists, join 3,700+ members inside the AI Profit Boardroom for the live calls, daily tutorials and done-for-you templates — or book a free SEO strategy session and get a personal agent-readiness plan first.
About the author
Julian Goldie is an SEO agency owner with 10+ years in SEO, 394K+ YouTube subscribers, a 100% Upwork job-success score, 75K+ community members across his groups, and a best-selling SEO book. He tests new AI tools the week they drop and publishes what actually works.
Watch the latest experiments on YouTube, learn AI SEO step by step inside the AI Profit Boardroom (3,700+ members, four live calls per week, daily tutorials, done-for-you templates and a 30-day roadmap), or book a free SEO strategy session to talk through your own setup.
Last updated September 2026. This is the living guide to how to use WebMCP — it gets updated as the tools change.
