WebMCP: What It Is and How ChatGPT Supports It

What is WebMCP: how a site hands its tools to AI agents

WebMCP is an experimental open standard that lets a website publish structured tools for AI agents. An agent is software that carries out a task for a person, such as searching a catalog or filling in a form. With WebMCP, the site spells out which actions it allows, and the agent uses them directly.

Without a standard like this, an agent has to guess how an interface works. WebMCP replaces the guess with a stated definition of how an agent can use the app.

OpenAI says agents then finish tasks faster, more accurately and more reliably. It describes the aim of its WebMCP Challenge as an app that gets meaningfully better when people and their agents use it together.

WebMCP support in ChatGPT: the in-app browser needs no setup

OpenAI says ChatGPT's in-app browser supports WebMCP out of the box. A developer who builds a WebMCP app can therefore test it there with no extra configuration.

OpenAI names two places to test: that browser and Google Chrome. Of the two, the in-app browser takes fewer steps, because Chrome needs WebMCP enabled first. If you want to watch an agent call your site's tools quickly, start in ChatGPT's in-app browser.

WebMCP in Chrome: the experimental flag and origin trial

Google Chrome works with WebMCP once the feature is enabled, either through an experimental flag or through the origin trial. An origin trial is Chrome's way of letting sites try a new feature before it ships to everyone.

OpenAI labels WebMCP experimental. The page names both Chrome routes as ways to test an app.

Chrome is also part of the Challenge itself. Google Chrome is listed among the companies providing additional prizes, and Sarah Drasner of Chrome was on the judging panel.

WebMCP examples and demos: OpenAI sample apps to start from

OpenAI provides five sample apps, each a WebMCP example. They cover:

• 3D modeling
• Collaborative writing
• A crossword builder
• Data exploration built on DuckDB-Wasm, a database engine that runs inside the browser
• An itinerary planner called Wandernote

The collaborative writing sample shows what a shared task looks like. A person and an agent write and revise one document together, and the agent can leave comments and respond under its own identity.

The data exploration sample works on data instead of text. Using DuckDB-Wasm, the agent queries and combines data alongside the person, and the two then create custom visualizations in the browser.

OpenAI WebMCP Challenge: $3,000 each for the top 10 entries

The WebMCP Challenge, OpenAI's contest for apps built on the standard, offered each of the top 10 submissions $3,000 in cash, a year of ChatGPT Pro, a Codex Micro keyboard and OpenAI swag. Shopify, Google Chrome, Netlify, Cloudflare, Vercel and Render were listed as providing additional prizes.

Registration and submissions opened on August 25, 2026. The submission deadline was September 3, 2026, so the contest is now closed.

Key dates, Pacific time

When	What
August 25, 2026, 12 p.m.	Registration and submissions open
August 25, 2026, 3 p.m.	Opening livestream
August 31, 2026, 11 a.m.	Office hours
September 3, 2026, 1 p.m.	Submission deadline
September 23, 2026	Winners announcement scheduled; the date might shift

The judges were Sarah Drasner (Chrome), Andrew Galloni (Cloudflare), Jude Gao (Vercel), Ilya Grigorik (Shopify), Sean Roberts (Netlify) and Justin Rushing (OpenAI), along with Alex Nahas, the creator of MCP-B. Entry did not require a new app. OpenAI encouraged new ones but also accepted entries that added WebMCP support to an existing app.

OpenAI's page scheduled the winners announcement for September 23, 2026 and said the date might shift. As of October 10, 2026, the page still shows that date and names no winners.