AI Catchup

OpenAI Developers Names 10 WebMCP Challenge Winners

By 5 min read

OpenAI Developers announced 10 WebMCP Challenge winners. The projects show websites exposing structured tools that agents can use alongside people, from editable floor plans and wedding seating to 3D scans, notebooks, and fantasy maps.

OpenAI Developers announced 10 winners of the WebMCP Challenge. The projects show what people and agents can build together when websites expose structured tools for agents to use. (OpenAI Developers on X; the challenge's "Meet the winners" update on Devpost)

WebMCP is an experimental open standard that lets websites expose structured tools directly to agents. OpenAI's challenge page frames the goal as making web apps meaningfully better when people and their agents use them together, rather than asking an agent to guess through a user interface. (OpenAI's WebMCP Challenge)

The 10 Projects OpenAI Highlighted

OpenAI's winner thread describes these projects, and the challenge's Devpost page lists the same 10 winners. The descriptions below stay close to the thread. (OpenAI Developers' winner thread; Devpost winners update)

  • MASIL uses WebMCP to help Korean elders practice calligraphy and play Janggi alongside an agent. (OpenAI Developers on X)
  • Alza turns a floor-plan photo into an editable 2D/3D model, with an agent checking the geometry as the user works. (OpenAI Developers on X)
  • ArchMorph lets a person design a home with an agent, walk through it in 3D, and edit and check the same live building model. (OpenAI Developers on X)
  • Aisle tackles wedding seating with an agent that respects guest relationships and pinned seats. (OpenAI Developers on X)
  • Roque Nights helps plan a night under the stars by comparing nights, finding visible targets, and proposing an observing plan for review. (OpenAI Developers on X)
  • Mandate turns a user's choice of CRM records, fields, and time limit into scoped WebMCP tools for an agent. (OpenAI Developers on X)
  • Observatory lets a person and an agent build fantasy maps with editable terrain, rooms, and labels. (OpenAI Developers on X)
  • Bouquet Studio turns requests such as “warmer” or “less formal” into flowers that a person and an agent can see and rearrange together. (OpenAI Developers on X)
  • Faraday lets an agent navigate 3D scans through WebMCP while the scan stays in the browser, for research and education. (OpenAI Developers on X)
  • JupyterLite WebMCP brings an agent into a live notebook to edit cells, run code, and review changes together. (OpenAI Developers on X)

Why the Examples Matter

The winning projects are not a single new model or API feature. They are demonstrations of a different integration boundary: the website exposes structured actions, and the agent can call those actions while the person stays in the loop. That makes the examples useful references for teams building agent-facing web applications, without implying that every project is a generally available product.

The examples also cover several interaction patterns:

  • Shared canvases: ArchMorph, Aisle, Bouquet Studio, and Observatory let people and agents change a visual artifact together.
  • Scoped or checked operations: Mandate gives the agent scoped tools, Alza has the agent check the geometry, and Roque Nights has the agent propose a plan for review.
  • In-browser work: Faraday keeps the scan in the browser and JupyterLite WebMCP works inside a browser-only notebook, while MASIL focuses on an assisted creative activity.

Those categories are an editorial grouping of the capabilities described in OpenAI's winner posts and the Devpost winners update. They are not additional claims about each project's implementation or availability. (OpenAI Developers' winner thread; Devpost winners update)

What Developers Can Take From WebMCP

OpenAI's challenge page says WebMCP lets a website define exactly how an agent can use an app. (OpenAI's WebMCP Challenge) Our reading of the winners is that the useful tools are actions that mean something at the application level: place a guest while respecting a constraint, edit a model, run a notebook cell, or produce a reviewable plan.

The page also says WebMCP can be tested in ChatGPT's in-app browser and in Google Chrome with WebMCP enabled through an experimental flag or origin trial. (OpenAI's WebMCP Challenge)

For teams evaluating the pattern, three questions are worth asking. They are ours, not OpenAI's:

  1. Which user-visible action should become a typed tool?
  2. Which constraints must the server enforce on every call?
  3. Where should a person review, correct, or approve the agent's changes?

The projects answer those questions in different domains, but the underlying idea is consistent: make the app's useful operations explicit so an agent can participate without replacing the human workflow.

Corrections and Updates

  • September 29, 2026: The FAQ answer on what kinds of projects won listed the sample apps on OpenAI's challenge page, which are not winners; it now lists the 10 winners. The three evaluation questions were attributed to the winner thread; they are editorial.

Sources

Read next

Related coverage

Frequently Asked Questions

What did OpenAI announce about the WebMCP Challenge?

OpenAI Developers announced 10 winning projects from the WebMCP Challenge. The examples pair websites with structured tools that agents can use while people review or direct the work.

What is WebMCP?

OpenAI describes WebMCP as an experimental open standard that lets websites expose structured tools agents can use directly, rather than making agents guess through a user interface.

What kinds of projects won the WebMCP Challenge?

The 10 winners are MASIL, Alza, ArchMorph, Aisle, Roque Nights, Mandate, Observatory, Bouquet Studio, Faraday, and JupyterLite WebMCP. They cover calligraphy and Janggi practice, floor plans, home design, wedding seating, stargazing, scoped CRM access, fantasy maps, bouquets, 3D scans, and notebooks.

Get the weekly AI Catchup

Tools, practices, and what matters, in your inbox every week.