Back to Gems of AI

Why Cursor's new MCP Apps completely change how we interact with code

Cursor just added MCP Apps, turning text-based AI chats into interactive dashboards and tools. Here is what this means for your workflow.

We have all hit the wall with text-based AI. You ask a language model to analyze a dataset or show you an architecture diagram, and it spits back a markdown table or an ASCII box. It works, but it feels incredibly limited.

Cursor just announced something that fixes this problem entirely. They call it MCP Apps (Model Context Protocol Apps). I spent the morning reading through the documentation, and I keep thinking about how many internal dashboards we suddenly do not need to build anymore.

Moving beyond plain text

The premise is straightforward. Instead of just returning text or code, your AI tools can now return fully interactive HTML interfaces right inside the Cursor chat.

Think about a standard workflow. You want to check your deployment metrics. Normally, you open a browser tab, log into AWS or Datadog, click through several menus, and find your graph.

With an MCP App, you type "show me the deployment metrics." The AI calls your monitoring tool and renders a live, interactive chart directly in the chat sidebar. You can zoom in or toggle data points. The interface lives right next to the conversation you are already having about the code.

How the mechanics actually work

Anthropic originally developed the Model Context Protocol to standardize how AI models talk to data sources. MCP Apps takes that foundation and adds a UI layer.

When the LLM decides to use an MCP App tool, the host fetches an HTML page. This page renders inside a secure sandbox. The clever part is the bidirectional communication. The app and the host talk to each other using a JSON-RPC protocol.

This means you can click a button in the custom UI, and that action gets sent back to the language model. The model can then trigger another tool, fetch fresh data, and push it back to your app. You get the state management and data flow of a real web application without leaving your editor.

Why you should care about this

You might wonder why you would not just build a quick React app and send a link. I asked myself the same thing.

The biggest advantage is context preservation. When you switch tabs, you lose your train of thought. Having the tool right there in the chat means you stay focused on the problem.

There is also a massive reduction in boilerplate. Building a standalone web app requires setting up authentication, routing, state management, and an API. MCP Apps piggyback on the connections your AI already has. If you want the app to schedule a meeting, it just tells the host to do it using the integrations you have already authorized.

Security is built into the foundation

Running third-party code in your editor sounds like a bad idea. Cursor handled this by forcing all MCP Apps into a sandboxed iframe.

These apps cannot access the parent page. They cannot read your cookies or touch local storage. All communication happens through a strict messaging API. This setup means you can safely run apps built by random developers without handing over the keys to your machine.

The end of throwaway dashboards

I genuinely do not know how many quick scripts and internal admin panels I have written over the years just to visualize some database rows. Most of them were abandoned after a month.

MCP Apps feel like the obvious replacement for all of that busywork. If you need a custom form to generate boilerplate code with 15 different options, you do not need an entire web app. You just need a UI that pops up when you ask for it.

I think we will see a huge ecosystem of these micro-apps emerge over the next few months. We are finally moving past the idea that AI has to talk to us like a command line interface.

Conclusion

Text chats are great for writing code, but they are terrible for exploring data. Cursor's implementation of MCP Apps solves a very specific, very annoying problem. Try building one this weekend. The barrier to entry is just standard web technologies, and the payoff is an editor that actually understands how you want to work.

Frequently Asked Questions

What are Cursor MCP apps?

MCP Apps are interactive, HTML-based user interfaces that render directly inside Cursor's chat window, allowing you to interact with data visually instead of just reading text.

Why do MCP apps matter for developers?

They eliminate the need to switch tabs or build separate web apps for simple internal tools. You get full bidirectional data flow between the AI and the custom UI without leaving your editor.

Are MCP apps secure?

Yes, they run inside a sandboxed iframe, which means they cannot access your parent page, steal cookies, or escape their isolated container.