Skip to content

MCP Apps: Interactive UIs Inside Claude and ChatGPT

MCP Apps is the first official MCP extension, live since January 26, 2026. It lets tools return interactive UI components that render directly in the conversation: dashboards, forms, visualizations, multi-step workflows, and more. This page covers what it is, how it works, client support, and the security model.

What MCP Apps enables

Before MCP Apps, a tool returned text. The user read a summary and asked follow-up questions to explore the data. With MCP Apps, a tool can return an interactive dashboard. The user filters by region, drills into a specific account, sorts a column, and exports a report without leaving the conversation. The model stays in the loop and sees what the user does.

Typical use cases:

  • Data exploration: a sales analytics tool returns an interactive dashboard
  • Configuration wizards: a deployment tool presents a form with dependent fields
  • Document review: a contract analysis tool displays the PDF inline with highlighted clauses
  • Real-time monitoring: a server health tool shows live metrics that update as systems change

How it works

MCP Apps relies on two MCP primitives:

  1. Tools with UI metadata. A tool declares a _meta.ui.resourceUri field pointing to a UI resource.
  2. UI resources. Server-side resources served via the ui:// scheme, containing bundled HTML and JavaScript.

Example tool declaration:

{
  name: "visualize_data",
  description: "Visualize data as an interactive chart",
  inputSchema: { /* ... */ },
  _meta: {
    ui: {
      resourceUri: "ui://charts/interactive"
    }
  }
}

The host fetches the resource, renders it in a sandboxed iframe, and enables bidirectional communication via JSON-RPC over postMessage. The UI can call server tools, update the model context, and send follow-up messages.

Client support

  • Claude: available on web and desktop since launch
  • ChatGPT: supported via the Apps SDK and MCP Apps compatibility
  • Visual Studio Code: supported
  • Goose: supported

Security model

  • Iframe sandboxing: all UI content runs in sandboxed iframes with restricted permissions
  • Pre-declared templates: hosts can review HTML content before rendering
  • Auditable messages: all UI-to-host communication goes through loggable JSON-RPC
  • User consent: hosts can require explicit approval for UI-initiated tool calls

MCP Apps and CorpusIQ

CorpusIQ uses MCP Apps to render verified business answers as interactive dashboards inside Claude and ChatGPT. The model computes the analysis, the CorpusIQ engine validates and reconciles the data, and the app renders the result with source citations and a reconciliation strip. See the runbook governance and response contract pages for the governance layer behind these answers.

Resources