Model Context Protocol gets its first official extension, changing what AI assistants can do. MCP Apps allow tools to return interactive user interfaces (dashboards, forms, visualizations, checkout flows) that are rendered directly within the conversation rather than as a text response.
Announced on January 26th, this extension represents a collaboration between Anthropic, OpenAI, and community maintainers. This addresses what the MCP team calls “one of the most requested features from the MCP community,” the ability to include interactive elements in AI responses that users can interact with without having to enter a separate prompt.
To understand why this is important, you need to understand what MCP is and how it is reshaping the AI tools ecosystem.
What is the Model Context Protocol?
The Model Context Protocol is an open standard Anthropic introduced in November 2024 to solve the fundamental problem that AI assistants were separated from the data and tools that people actually use. All integrations connecting AI to calendars, files, and business software required custom implementation.
MCP standardizes these connections. Think of it like USB-C for AI applications. Just as USB-C provides a universal way to connect devices to peripherals, MCP provides a universal way to connect AI models to external systems.
This protocol follows a client/server architecture. An MCP host (such as Claude Desktop or ChatGPT) connects to an MCP server, which is a lightweight program that exposes specific functionality. MCP servers may provide access to Google Calendar, your company’s database, or specialized tools such as Figma.
What makes MCP powerful is that the AI model becomes an active participant in the data rather than a passive receiver. Models can invoke tools at runtime through protocols to perform actions rather than simply describing the actions to be performed.
This standard quickly became popular. OpenAI officially adopted MCP in March 2025. In December, Anthropic donated the protocol to the Linux Foundation’s Agentic AI Foundation, whose members include Google, Microsoft, and AWS. The same open standards approach that Anthropic used for its skills framework is now shaping how the entire industry builds AI integrations.
What the MCP app adds
Previously, the MCP tool returned text. When you ask AI to check your calendar, it will explain your schedule in words. If you ask them to analyze your data, they will summarize the results in a paragraph. The AI had access to the tool, but the output was always text rendered in the chat window.
MCP Apps changes this. Tools can now return an HTML interface to render as an interactive element within a conversation. The calendar tool may display an actual calendar grid where you can click on dates. Data analysis tools may display graphs that you can hover over to see more details. Shopping tools may include a checkout form that you can fill out directly.
The technical implementation uses sandboxed iframes for security. The MCP server declares UI templates in advance, and the client (Claude, ChatGPT, or other host) renders them in an isolated environment that prevents malicious code execution.
This experience transforms the AI from a conversation partner explaining actions to an interface layer presenting actionable controls. The model stays in the loop, checking user behavior and responding accordingly, while the UI handles things that text can’t handle, such as live updates, native media viewers, persistent state, and direct manipulation.
why is this important
Let’s consider the actual difference. Without MCP Apps, exploring the data requires a repeated prompt to “Show sales by region.” “Then narrow it down to Q4.” “Sort by revenue.” Each interaction means entering a new prompt and waiting for a text response.
When using the MCP app, the AI returns an interactive data table. Click the column headers to sort. Drag the slider to filter the date range. Hover your mouse over the value to see details. The AI can monitor these interactions and respond with, “We’ve noticed you’re focusing on the Northeast. We’ll do a more detailed analysis,” but exploration is done through direct interaction rather than conversation.
This closes the gap that has limited AI assistants since the inception of ChatGPT. Adobe’s integration into ChatGPT hinted at what is possible when AI can present visual interfaces. MCP Apps standardizes its functionality so any developer can build it.
Release partners and availability
Anthropic has rolled out support for MCP apps on Claude for Pro, Max, Team, and Enterprise subscribers. Initial integrations will be provided by launch partners including Amplitude, Asana, Box, Canva, Clay, Figma, Hex, monday.com, and Slack. Salesforce integration is coming soon.
Real results: Users can build project timelines in Asana, draft formatted Slack messages, create and edit Figma diagrams, and manage Box files, all from within Claude’s chat interface. Each tool displays a native UI rather than forcing the user to describe what they want in text.
For developers, Anthropic has published an ext-apps repository containing the SDK and working samples. The reference implementation includes 3D visualization, interactive maps, PDF display, real-time system monitoring dashboard, and musical notation with Three.js. An open specification means developers can build MCP apps that work with any client that supports extensions.
big picture
MCP Apps continues Anthropic’s strategy of building industry infrastructure as open standards. The company currently offers MCP for tool connectivity, agent skills for feature customization, and MCP apps for conversational interfaces, each of which is released openly rather than as a standalone feature.
This approach reverses the dynamics of traditional software. Instead of an app containing AI functionality, the AI becomes an interface for accessing the app. MCP Apps enriches the interface by allowing tools to display visual controls rather than just textual descriptions.
For users, the immediate benefit is a smoother workflow. Actions that require switching between apps or providing detailed prompts can be performed by clicking and dragging. For developers, MCP Apps offers a new distribution channel. Build your interactive tool once and it will work within the AI assistant that supports its extensions.
This extension is operationally ready as of January 26th. Whether MCP Apps become as ubiquitous as MCP itself will depend on how quickly developers can build compelling implementations, and how well sandboxed iframe architectures address the security challenges of running arbitrary web interfaces within AI conversations.
