MCP-UI

Ido Salomon and Liad Yosef, both at Monday.com, built MCP-UI as an open protocol and SDK for returning interactive web components over MCP rather than plain text. It reuses the embedded resources of the Model Context Protocol specification: a server can answer with an external URL, raw HTML, or remote DOM, rendered inside sandboxed iframes that reach the host only through post messages and cannot touch its origin, cookies, or memory.

The recurring argument is about the value chain. When Shopify, Airbnb, or Amazon answer an agent with text alone, the chat decides how the response looks, and the provider loses its visual identity along with years of refined interface work. MCP-UI lets each application send chunks of its own UI instead.

Yosef's communication spectrum sets four levels of responsibility over user actions: notify, tool call, prompt, intent. His sprint demo walks all of them. A Monday widget surfaces stuck tasks, a click sends a prompt to the agent, the agent reads Gmail through a separate MCP server and finds that Sarah is ill while Jordan knows the code, and an intent message triggers the reassignment. Monday built no Gmail integration for any of it.

By the October 2025 talk: Shopify deployed across its stores, Postman, Goose, Libra chat, Hugging Face exposing every space, 11 Labs, and community SDKs in Ruby, Python, TypeScript and Go. Andrew Harvard of Block discussed the technology on the Wild Goose Case podcast. SSO stays unsolved, and native clients that support neither iframes nor webviews remain an open design question.