OpenAI names ten winners of its WebMCP agent challenge
OpenAI's WebMCP Challenge winners include an agent that helps design and walk through a home in 3D and one that turns a floor-plan photo into an editable model.
OpenAI announced the ten winners of its WebMCP Challenge on Monday. It is a Devpost competition for projects where a website exposes tools an AI agent can call directly, instead of clicking through the page like a person would.
Among the winners: ArchMorph, built by the developer xmusfk, lets a person design a home alongside an agent, then walk through the result in 3D. The agent edits and checks the same live building model, OpenAI said. Alza, built by a developer using the handle Elioz 404, turns a photo of a floor plan into an editable 2D and 3D model. An agent verifies the geometry as changes are made.
https://x.com/OpenAIDevs/status/2104617557345079375
https://x.com/OpenAIDevs/status/2104617512197611889
A third winner, MASIL, was built by the developer SonSangjoon. It uses the same WebMCP tools to let Korean elders practice calligraphy and play the board game Janggi alongside an agent, OpenAI said. OpenAI described the ten projects as showing what "people and agents can build together" when a site exposes structured tools rather than only a page to browse. None of the winning demos has been reviewed independently beyond its own account.
https://x.com/OpenAIDevs/status/2104617398980809040
OpenAI's post naming the winners had drawn more than 500 likes by Monday afternoon. In the replies, the developer BigP_dev called the approach "way better than watching it hunt for the submit button." That is a comparison to agents that must click through web pages built only for people.