Build interactive AI apps for MCP Apps and ChatGPT from a single codebase.

Quick Start

npx @mcp-apps-kit/create-app@latest my-app
cd my-app && npm run dev

Features

• 🔌 Single codebase — Define tools and UI once, deploy to MCP Apps & ChatGPT
• 🎯 Type-safe — Full TypeScript inference for inputs, outputs, and UI access
• ⚛️ React bindings — First-class hooks, context, and component UIs
• 🔐 OAuth 2.1 — JWT validation and JWKS discovery built-in
• 🧪 Testing utilities — Property-based testing, UI mocks, and LLM evaluation
• 📦 Flexible deployment — Express server, serverless, or stdio
• 🔀 API versioning — Expose multiple versions from a single app

Table of Contents

• Packages
• Install
• Usage
• Deployment
• Platform Support
• Examples
• API
• Contributing

---

Background

MCP AppsKit is a TypeScript framework for building interactive applications with shared tool and UI definitions. MCP Apps enables servers to deliver interactive UIs to MCP hosts. OpenAI Apps SDK powers apps inside ChatGPT. This framework abstracts both protocols, providing a server runtime, client SDK, React bindings, and a scaffolding CLI.

Best for: One codebase serving MCP Apps + ChatGPT, type-safe tools with Zod, tool-to-UI rendering patterns.

May not fit: Single protocol with low-level control needs, or custom transport/session behavior beyond the built-in server.

Packages

Package	Description
@mcp-apps-kit/core	Server-side framework
@mcp-apps-kit/ui	Client-side SDK (vanilla JS)
@mcp-apps-kit/ui-react	React bindings
@mcp-apps-kit/ui-react-builder	Build tool for React component UIs
@mcp-apps-kit/create-app	CLI scaffolding tool
@mcp-apps-kit/testing	Testing utilities for MCP apps

Compatibility

• Node.js: >= 18 (runtime packages), >= 20 (@mcp-apps-kit/create-app CLI and monorepo development)
• React: 18.x or 19.x (peer dependency)
• Zod: ^4.0.0 required

Install

# Server framework
npm install @mcp-apps-kit/core zod

# React UI (includes @mcp-apps-kit/ui)
npm install @mcp-apps-kit/ui-react

See Packages for all available packages.

Usage

Server

import { createApp, defineTool, defineUI } from "@mcp-apps-kit/core";
import { z } from "zod";

// Define a UI resource for the tool
const greetingUI = defineUI({
  name: "Greeting Widget",
  html: "./ui/dist/greeting.html",
  prefersBorder: true,
});

const app = createApp({
  name: "my-app",
  version: "1.0.0",
  tools: {
    greet: defineTool({
      title: "Greet User",
      description: "Generate a personalized greeting",
      input: z.object({
        name: z.string().describe("Name to greet"),
      }),
      output: z.object({
        message: z.string(),
        timestamp: z.string(),
      }),
      ui: greetingUI,
      handler: async ({ name }) => ({
        message: `Hello, ${name}!`,
        timestamp: new Date().toISOString(),
      }),
    }),
  },
});

await app.start({ port: 3000 });

React UI

Wrap your app with AppsProvider and use hooks to access tool results:

// App.tsx
import { AppsProvider } from "@mcp-apps-kit/ui-react";
import { GreetingWidget } from "./GreetingWidget";

export function App() {
  return (
    <AppsProvider>
      <GreetingWidget />
    </AppsProvider>
  );
}

// GreetingWidget.tsx
import { useToolResult, useHostContext } from "@mcp-apps-kit/ui-react";

interface GreetingResult {
  message: string;
  timestamp: string;
}

export function GreetingWidget() {
  const result = useToolResult<GreetingResult>();
  const { theme } = useHostContext();

  if (!result) {
    return <div>Loading...</div>;
  }

  return (
    <div style={{ background: theme === "dark" ? "#1a1a1a" : "#fff" }}>
      <h2>{result.message}</h2>
      <p>Generated at: {new Date(result.timestamp).toLocaleString()}</p>
    </div>
  );
}

For complete examples including API versioning, React component UIs, and advanced patterns, see the Examples section.

Deployment

Express (default)

await app.start({ port: 3000 });

Custom Express

import express from "express";
const expressApp = express();
expressApp.use("/mcp", app.handler());
expressApp.listen(3000);

Stdio

await app.start({ transport: "stdio" });

Serverless

export default {
  async fetch(request) {
    return app.handleRequest(request);
  },
};

Platform Support

Feature	MCP Apps	ChatGPT Apps
Tool Calling	Yes	Yes
Structured Data	Yes	Yes
OAuth 2.1 Auth	Yes	Yes
Theme Support	Yes	Yes
Display Modes	Yes	Yes
Persisted State	No	Yes
File Upload	No	Yes
Tool Cancellation	Yes	No
Debug Logging	Yes	Yes

Examples

Kanban demo repository (tool calling, React widgets, plugins, middleware, events):

git clone https://github.com/AndurilCode/kanban-mcp-example.git
cd kanban-mcp-example
npm install
npm run dev

Local examples:

• examples/minimal - minimal server with API versioning
• examples/weather-app - weather tools with React UI (Open-Meteo API)
• examples/restaurant-finder - search functionality with UI

API

📖 API Reference - Full TypeDoc-generated documentation

Detailed package documentation:

• packages/core/README.md
• packages/ui/README.md
• packages/ui-react/README.md
• packages/ui-react-builder/README.md
• packages/create-app/README.md
• packages/testing/README.md

Contributing

See CONTRIBUTING.md for development setup and guidelines. Issues and pull requests are welcome.

License

MIT