MCP Apps: Why Interactive UIs Inside AI Conversations Will Change How We Build Software

MCP Apps, the Model Context Protocol's most important upgrade, are set to transform how developers, businesses, and users experience human-AI interaction.

MCP Apps End the Copy-Paste Workflow Era

For the past two years, AI assistants have been remarkably powerful — and remarkably limited. You could ask Claude to analyze your sales data, but the results came back as text. You could ask ChatGPT to help plan a sprint, but you'd need to manually transfer everything to Jira. The AI understood your intent perfectly, then delivered the answer in the wrong format.

On January 26, 2026, the Model Context Protocol shipped its first official extension: MCP Apps. And it might be the most important development in AI tooling since MCP itself launched in late 2024.

MCP Apps, the first official extension to Anthropic's Model Context Protocol, lets tools return interactive UI components that render directly inside the AI conversation. Not links to external dashboards. Not screenshots. Actual, functional interfaces — with buttons, sliders, forms, real-time data, and bidirectional communication between the UI and the AI model.

The implications of MCP Apps go far beyond "chat but prettier."

What MCP Apps Are and How MCP Apps Work

At its core, MCP Apps is deceptively simple. When an MCP server defines a tool, it can now include a _meta.ui.resourceUri field pointing to a bundled HTML/JavaScript application. When the tool is invoked, the host (Claude, ChatGPT, VS Code, Goose) renders that application in a sandboxed iframe within the conversation.

The key innovation isn't the iframe — it's the bidirectional communication layer. The embedded app can:

• Receive tool results from the AI model
• Call server tools back (fetching more data, triggering actions)
• Update the model's context (telling the AI what the user just clicked, selected, or configured)
• Open links in the user's browser
• Send follow-up messages to drive the conversation forward

MCP Apps create a feedback loop: the AI suggests, the user interacts with the UI, the UI tells the AI what happened, and the AI adjusts. It's the first time we've had genuine human-AI-UI collaboration in a single interface.

The Launch Partners Tell the Story

When MCP Apps launched, nine integrations shipped on day one: Amplitude, Asana, Box, Canva, Clay, Figma, Hex, Monday.com, and Slack — with Salesforce close behind. By week two, over 75 apps were available in Claude alone.

These MCP Apps integrations aren't toy demos. Consider the real workflows:

• Figma in Claude: Discuss a feature with your team, then create and iterate on diagrams in the same conversation. The AI sees your changes and suggests improvements in real-time.
• Amplitude in ChatGPT: Ask about pipeline metrics and get an interactive dashboard. Filter by region, drill into accounts, export reports — all without leaving the chat.
• Asana in Goose: Plan a sprint conversationally while adjusting timelines, reassigning tasks, and updating dependencies inline.

The pattern is consistent: eliminate the context switch. The cognitive overhead of bouncing between AI assistant and specialized tool disappears entirely.

The Protocol Stack: MCP Apps in Context

MCP Apps doesn't exist in isolation. It's part of an emerging agentic protocol stack that's defining how AI systems interact with tools, each other, and humans:

MCP Apps occupies a unique position: it's the UI layer of the most widely adopted AI integration protocol. With 97 million monthly SDK downloads and 10,000+ active servers, MCP has already won the tools layer. MCP Apps extends that dominance into the interface layer.

CopilotKit's AG-UI protocol is complementary, not competing — it provides the runtime event/state transport, while MCP Apps provides the UI specification. In fact, CopilotKit announced MCP Apps support on the same day, letting developers bring MCP App UIs into their own applications.

Why MCP Apps Change Everything for Developers

1. Write Once, Run Everywhere (For Real This Time)

Build an MCP App and it works in Claude, ChatGPT, VS Code, Goose, JetBrains IDEs, and any future client that supports the protocol. One codebase. Every major AI platform.

MCP Apps provide the first genuine cross-platform UI portability for AI tools. Before MCP Apps, building a ChatGPT plugin, a Claude integration, and a VS Code extension meant three separate codebases. Now it's one.

2. The AI Sees What the User Does

This is the subtle but revolutionary part. When a user interacts with an MCP App — clicks a button, adjusts a slider, selects an option — the app can call updateModelContext() to inform the AI. The model doesn't just generate a UI and walk away. It stays in the loop.

// When user selects a configuration option
await app.updateModelContext({
  content: [{ 
    type: "text", 
    text: "User selected production environment with enhanced security" 
  }]
});

This enables stateful, context-aware interactions that were previously impossible. The AI can offer increasingly relevant suggestions based on what the user is actually doing in the UI.

3. Tools Become Products

Before MCP Apps, an MCP server was invisible infrastructure. Now it can have a face. Developers can build tools that don't just process data — they present experiences. An analytics tool doesn't just return numbers; it presents an interactive dashboard. A project estimator doesn't just calculate costs; it lets you explore scenarios with sliders and toggles.

MCP Apps transform MCP servers from backend utilities into distributable products with real user interfaces.

The Security Model: Thoughtful, Not Perfect

Running third-party UI code inside an AI assistant is inherently risky. MCP Apps addresses this through layered defenses:

• Iframe sandboxing: UI runs in restricted iframes — no parent window access, no arbitrary network requests
• Pre-declared templates: Hosts review HTML before rendering, preventing dynamic code injection
• Auditable JSON-RPC: All UI-to-host communication is loggable — no backchannel
• User consent: Hosts can require approval for UI-initiated tool calls

For enterprises, this means maintaining allowlists of approved servers, auditing message logs, and implementing least-privilege access controls. The specification provides primitives; implementation is the organization's responsibility.

What MCP Apps Mean for Business

The New Distribution Channel

MCP Apps creates a distribution channel that didn't exist before. When someone connects an MCP server, they get access to your tools and your UI. Your interactive experiences reach users wherever they work with AI — which, increasingly, is everywhere.

For SaaS companies, this means your product can live inside Claude, ChatGPT, and VS Code simultaneously without maintaining three separate integrations. For agencies and studios, it means building MCP Apps becomes a service offering — one that's in early-stage demand with almost zero competition.

The End of the App Switcher

Andrew Harvard from Block frames it well: "The future centers on users navigating through one trusted agent rather than context-switching between fragmented experiences."

If this plays out, the AI assistant becomes the primary interface for knowledge work. Individual apps become pluggable components that surface when needed and recede when they're not. You don't open Figma — you tell Claude you need a diagram, and Figma appears.

This is a fundamentally different model than Microsoft 365 Copilot or Google Gemini Workspace, which embed AI into existing productivity suites. MCP Apps inverts the relationship: apps plug into the AI, not the other way around.

For Companies Evaluating AI Strategy

If you're building a product that integrates with AI:

1. Build an MCP server now — it's the most widely adopted integration standard
2. Add MCP Apps for interactive experiences — your competitors probably haven't yet
3. Think UI-first — the text-only era of AI tools is ending
4. Design for context — your app will live alongside other tools in a conversation; make it compositional

If you're adopting AI tools for your team:

1. Evaluate MCP-compatible platforms — they'll give you the broadest tool ecosystem
2. Audit your tool sprawl — MCP Apps can consolidate multi-tool workflows into single-conversation experiences
3. Train for conversational-first workflows — your team's interaction patterns are about to change

What We're Building

At Context Studios, we've been building MCP servers since the protocol launched. Our content automation server has 45+ tools across seven domains — blog publishing, social media, SEO, image generation, research, and knowledge management. We recently upgraded our MCP Apps to include:

• An interactive project estimator with sliders for scope adjustment, feature toggles, architecture diagrams, and built-in lead capture
• A content discovery hub with full article rendering, tag-based filtering, search, and language switching

These aren't demos — they're production tools that run inside Claude and ChatGPT, providing real value to anyone who connects our server.

The opportunity is massive and early. Almost nobody has production MCP Apps yet. The ecosystem is wide open for developers who move fast.

The Bigger Picture: What Comes Next for MCP Apps

MCP Apps is the first official MCP extension, but it won't be the last. The specification establishes a pattern for extending the protocol beyond its core primitives. As Clare Liguori from AWS noted: "The ability to render dynamic interfaces directly in conversation makes it easier to leverage MCP server capabilities in practical ways."

Here's what we think comes next:

1. Generative UI: AI models that don't just invoke pre-built UIs but generate custom interfaces on the fly. The building blocks are already here in A2UI and AG-UI.
2. Multi-agent App orchestration: Agent Teams (Opus 4.6) + MCP Apps = multiple agents coordinating while presenting unified interfaces to users.
3. Marketplace dynamics: As MCP server registries mature, MCP Apps with great UIs will have a distribution advantage — users prefer tools they can see and interact with.
4. Enterprise workflow consolidation: The "one agent, many tools" pattern will replace the "many apps, many tabs" pattern for knowledge workers within 2-3 years.

The shift is from text-first to experience-first AI interactions. MCP Apps is the first real infrastructure for making that happen.

Getting Started

The ext-apps repository includes working examples: 3D visualizations, interactive maps, document viewers, real-time dashboards, and music notation. The SDK is @modelcontextprotocol/ext-apps on npm.

If you're a developer: pick an example close to what you're building and start there. The development loop is straightforward — declare tools with UI metadata, serve bundled HTML via ui://, use the App class for communication.

If you're a business: talk to your development team about MCP integration. The window for early-mover advantage is open right now.

---

Context Studios is an AI-native development studio based in Berlin. We build MCP servers, AI agents, and modern web applications. If you're looking to build an MCP App or integrate AI into your product, let's talk.

FAQ

What's the difference between MCP and MCP Apps?

MCP (Model Context Protocol) is the base protocol for connecting AI models to external tools and data. MCP Apps is the first official extension that adds interactive UI capabilities — tools can now return visual interfaces, not just text.

Which AI platforms support MCP Apps?

As of February 2026: Claude (web + desktop), ChatGPT, VS Code, Goose, and JetBrains IDEs are exploring support. The standard is cross-platform by design.

Do I need to build separate UIs for each AI platform?

No. That's the key advantage — build one MCP App and it works across all compatible clients. The host handles rendering and sandboxing.

Is it safe to run third-party UIs inside AI conversations?

MCP Apps uses iframe sandboxing, pre-declared templates, auditable JSON-RPC communication, and user consent requirements. Enterprise users should maintain server allowlists and audit logs.

How does MCP Apps compare to ChatGPT plugins?

MCP Apps is an open standard governed by the Agentic AI Foundation (Linux Foundation), not a proprietary platform. It works across multiple AI clients, not just one.

Can MCP Apps access my data?

Only what you explicitly grant. MCP Apps inherits MCP's security model: users must consent to data access, and hosts control what information is shared with servers.