---
type: "Talk"
title: "MCP Apps: UI in AI"
description: "MCP Apps: UI in AI"
resource: "https://aurelijus.banelis.lt/en/talks/mcp-apps-ui-in-ai/"
timestamp: "2026-07-31"
tags:
  - "ai"
  - "mcp"
  - "ui"
  - "llm"
  - "NoTrollsAllowed"
lang: "en"
date: "2026-07-31"
---


### UI in AI history

* The first wave of AI adoption focused on embedded chat bots inside their UI. Some
  companies even tried [custom GPTs](https://www.ingka.com/newsroom/ikea-launches-new-ai-powered-assistant-in-openai-gpt-store/).
* Then came MCP, Skills, ACP and control switched to the consumer. But UI was mostly
  controlled by the apps of AI giants (ChatGPT, Google, Anthropic, X)

### Capabilities for UI innovation

As an AI-driven company, it is expensive to compete on cheaper service alone, or to
train your own model.

Luckily, in the [latest MCP specification](https://modelcontextprotocol.io/specification/2026-07-28)
comes [**MCP Apps**](https://apps.extensions.modelcontextprotocol.io/api/).

It is basically an HTML page inside your AI tool. Implementation is as simple as
returning a resource via a `ui://` path.

### Why should I care?

* While adoption is still growing, MCP Apps could be the [next focus](https://apps.extensions.modelcontextprotocol.io/api/media/SKILL-3.md)
  for Frontend software developers
* While generic AI is mostly competing via text output, I am betting for smaller
  companies to leverage competition over UX.