Skip to content
Cloudflare Docs

Build an Interactive ChatGPT App

Deploy your first ChatGPT App

This guide will show you how to build and deploy an interactive ChatGPT App on Cloudflare Workers that can:

  • Render rich, interactive UI widgets directly in ChatGPT conversations
  • Maintain real-time, multi-user state using Durable Objects
  • Enable bidirectional communication between your app and ChatGPT
  • Build multiplayer experiences that run entirely within ChatGPT

You will build a real-time multiplayer chess game that demonstrates these capabilities. Players can start or join games, make moves on an interactive chessboard, and even ask ChatGPT for strategic advice—all without leaving the conversation.

Your ChatGPT App will use the Model Context Protocol (MCP) to expose tools and UI resources that ChatGPT can invoke on your behalf.

You can view the full code for this example here ↗.

Prerequisites

Before you begin, you will need:

1. Enable ChatGPT Developer Mode

To use ChatGPT Apps (also called connectors), you need to enable developer mode:

  1. Open ChatGPT ↗.
  2. Go to Settings > Apps & Connectors > Advanced Settings
  3. Toggle Developer mode ON

Once enabled, you will be able to install custom apps during development and testing.

2. Create your ChatGPT App project

  1. Create a new project for your Chess App:
Terminal window
npm create cloudflare@latest -- my-chess-app
  1. Navigate into your project:
Terminal window
cd my-chess-app
  1. Install the required dependencies:
Terminal window
npm install agents @modelcontextprotocol/sdk chess.js react react-dom react-chessboard
  1. Install development dependencies:
Terminal window
npm install -D @cloudflare/vite-plugin @vitejs/plugin-react vite vite-plugin-singlefile @types/react @types/react-dom

3. Configure your project

  1. Update your wrangler.jsonc to configure Durable Objects and assets:
{
  "name": "my-chess-app",
  "main": "src/index.ts",
  "compatibility_date": "2025-01-01",
  "compatibility_flags": ["nodejs_compat"],
  "durable_objects": {
    "bindings": [
      {
        "name": "CHESS",
        "class_name": "ChessGame"
      }
    ]
  },
  "migrations": [
    {
      "tag": "v1",
      "new_sqlite_classes": ["ChessGame"]
    }
  ],
  "assets": {
    "directory": "dist",
    "binding": "ASSETS"
  }
}
  1. Create a vite.config.ts for building your React UI:
TypeScript
import { cloudflare } from "@cloudflare/vite-plugin";
import react from "@vitejs/plugin-react";
import { defineConfig } from "vite";
import { viteSingleFile } from "vite-plugin-singlefile";


export default defineConfig({
  plugins: [react(), cloudflare(), viteSingleFile()],
  build: {
    minify: false
  }
});
  1. Update your package.json scripts:
{
  "scripts": {
    "dev": "vite",
    "build": "vite build",
    "deploy": "vite build && wrangler deploy"
  }
}

4. Create the Chess game engine

  1. Create the game logic using Durable Objects at src/chess.tsx:
import { Agent, callable, getCurrentAgent } from "agents";
import { Chess } from "chess.js";


type Color = "w" | "b";


type ConnectionState = {
  playerId: string;
};


export type State = {
  board: string;
  players: { w?: string; b?: string };
  status: "waiting" | "active" | "mate" | "draw" | "resigned";
  winner?: Color;
  lastSan?: string;
};


export class ChessGame extends Agent<Env, State> {
  initialState: State = {
    board: new Chess().fen(),
    players: {},
    status: "waiting"
  };


  game = new Chess();


  constructor(ctx: DurableObjectState, public env: Env) {
    super(ctx, env);
    this.game.load(this.state.board);
  }


  private colorOf(playerId: string): Color | undefined {
    const { players } = this.state;
    if (players.w === playerId) return "w";
    if (players.b === playerId) return "b";
    return undefined;
  }


  @callable()
  join(params: { playerId: string; preferred?: Color | "any" }) {
    const { playerId, preferred = "any" } = params;
    const { connection } = getCurrentAgent();
    if (!connection) throw new Error("Not connected");


    connection.setState({ playerId });
    const s = this.state;


    // Already seated? Return seat
    const already = this.colorOf(playerId);
    if (already) {
      return { ok: true, role: already as Color, state: s };
    }


    // Choose a seat
    const free: Color[] = (["w", "b"] as const).filter((c) => !s.players[c]);
    if (free.length === 0) {
      return { ok: true, role: "spectator" as const, state: s };
    }


    let seat: Color = free[0];
    if (preferred === "w" && free.includes("w")) seat = "w";
    if (preferred === "b" && free.includes("b")) seat = "b";


    s.players[seat] = playerId;
    s.status = s.players.w && s.players.b ? "active" : "waiting";
    this.setState(s);
    return { ok: true, role: seat, state: s };
  }


  @callable()
  move(
    move: { from: string; to: string; promotion?: string },
    expectedFen?: string
  ) {
    if (this.state.status === "waiting") {
      return {
        ok: false,
        reason: "not-in-game",
        fen: this.game.fen(),
        status: this.state.status
      };
    }


    const { connection } = getCurrentAgent();
    if (!connection) throw new Error("Not connected");
    const { playerId } = connection.state as ConnectionState;


    const seat = this.colorOf(playerId);
    if (!seat) {
      return {
        ok: false,
        reason: "not-in-game",
        fen: this.game.fen(),
        status: this.state.status
      };
    }


    if (seat !== this.game.turn()) {
      return {
        ok: false,
        reason: "not-your-turn",
        fen: this.game.fen(),
        status: this.state.status
      };
    }


    // Optimistic sync guard
    if (expectedFen && expectedFen !== this.game.fen()) {
      return {
        ok: false,
        reason: "stale",
        fen: this.game.fen(),
        status: this.state.status
      };
    }


    const res = this.game.move(move);
    if (!res) {
      return {
        ok: false,
        reason: "illegal",
        fen: this.game.fen(),
        status: this.state.status
      };
    }


    const fen = this.game.fen();
    let status: State["status"] = "active";
    if (this.game.isCheckmate()) status = "mate";
    else if (this.game.isDraw()) status = "draw";


    this.setState({
      ...this.state,
      board: fen,
      lastSan: res.san,
      status,
      winner: status === "mate" ? (this.game.turn() === "w" ? "b" : "w") : undefined
    });


    return { ok: true, fen, san: res.san, status };
  }


  @callable()
  resign() {
    const { connection } = getCurrentAgent();
    if (!connection) throw new Error("Not connected");
    const { playerId } = connection.state as ConnectionState;


    const seat = this.colorOf(playerId);
    if (!seat) return { ok: false, reason: "not-in-game", state: this.state };


    const winner = seat === "w" ? "b" : "w";
    this.setState({ ...this.state, status: "resigned", winner });
    return { ok: true, state: this.state };
  }
}

5. Create the MCP server and UI resource

  1. Create your main worker at src/index.ts:
TypeScript
import { createMcpHandler } from "agents/mcp";
import { routeAgentRequest } from "agents";
import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
import { env } from "cloudflare:workers";


const getWidgetHtml = async (host: string) => {
  let html = await (await env.ASSETS.fetch("http://localhost/")).text();
  html = html.replace(
    "<!--RUNTIME_CONFIG-->",
    `<script>window.HOST = \`${host}\`;</script>`
  );
  return html;
};


const server = new McpServer({ name: "Chess", version: "v1.0.0" });


// Register a UI resource that ChatGPT can render
server.registerResource(
  "chess",
  "ui://widget/index.html",
  {},
  async (_uri, extra) => {
    return {
      contents: [
        {
          uri: "ui://widget/index.html",
          mimeType: "text/html+skybridge",
          text: await getWidgetHtml(extra.requestInfo?.headers.host as string)
        }
      ]
    };
  }
);


// Register a tool that ChatGPT can call to render the UI
server.registerTool(
  "playChess",
  {
    title: "Renders a chess game menu, ready to start or join a game.",
    annotations: { readOnlyHint: true },
    _meta: {
      "openai/outputTemplate": "ui://widget/index.html",
      "openai/toolInvocation/invoking": "Opening chess widget",
      "openai/toolInvocation/invoked": "Chess widget opened"
    }
  },
  async (_, _extra) => {
    return {
      content: [{ type: "text", text: "Successfully rendered chess game menu" }]
    };
  }
);


const mcpHandler = createMcpHandler(server);


export default {
  async fetch(req: Request, env: Env, ctx: ExecutionContext) {
    const url = new URL(req.url);
    if (url.pathname.startsWith("/mcp")) return mcpHandler(req, env, ctx);


    return (
      (await routeAgentRequest(req, env)) ??
      new Response("Not found", { status: 404 })
    );
  }
};


export { ChessGame } from "./chess";

6. Build the React UI

  1. Create the HTML entry point at index.html:
<!doctype html>
<html>
  <head>
    <!--RUNTIME_CONFIG-->
  </head>
  <body>
    <div id="root" style="font-family: verdana"></div>
    <script type="module" src="/src/app.tsx"></script>
  </body>
</html>
  1. Create the React app at src/app.tsx:
import { useEffect, useRef, useState } from "react";
import { useAgent } from "agents/react";
import { createRoot } from "react-dom/client";
import { Chess, type Square } from "chess.js";
import { Chessboard, type PieceDropHandlerArgs } from "react-chessboard";
import type { State as ServerState } from "./chess";


function usePlayerId() {
  const [pid] = useState(() => {
    const existing = localStorage.getItem("playerId");
    if (existing) return existing;
    const id = crypto.randomUUID();
    localStorage.setItem("playerId", id);
    return id;
  });
  return pid;
}


function App() {
  const playerId = usePlayerId();
  const [gameId, setGameId] = useState<string | null>(null);
  const [gameIdInput, setGameIdInput] = useState("");
  const [menuError, setMenuError] = useState<string | null>(null);


  const gameRef = useRef(new Chess());
  const [fen, setFen] = useState(gameRef.current.fen());
  const [myColor, setMyColor] = useState<"w" | "b" | "spectator">("spectator");
  const [pending, setPending] = useState(false);
  const [serverState, setServerState] = useState<ServerState | null>(null);
  const [joined, setJoined] = useState(false);


  const host = window.HOST ?? "http://localhost:5173/";


  const { stub } = useAgent<ServerState>({
    host,
    name: gameId ?? "__lobby__",
    agent: "chess",
    onStateUpdate: (s) => {
      if (!gameId) return;
      gameRef.current.load(s.board);
      setFen(s.board);
      setServerState(s);
    }
  });


  useEffect(() => {
    if (!gameId || joined) return;


    (async () => {
      try {
        const res = await stub.join({ playerId, preferred: "any" });
        if (!res?.ok) return;


        setMyColor(res.role);
        gameRef.current.load(res.state.board);
        setFen(res.state.board);
        setServerState(res.state);
        setJoined(true);
      } catch (error) {
        console.error("Failed to join game", error);
      }
    })();
  }, [playerId, gameId, stub, joined]);


  async function handleStartNewGame() {
    const newId = crypto.randomUUID();
    setGameId(newId);
    setGameIdInput(newId);
    setMenuError(null);
    setJoined(false);
  }


  async function handleJoinGame() {
    const trimmed = gameIdInput.trim();
    if (!trimmed) {
      setMenuError("Enter a game ID to join.");
      return;
    }
    setGameId(trimmed);
    setMenuError(null);
    setJoined(false);
  }


  const handleHelpClick = () => {
    window.openai?.sendFollowUpMessage?.({
      prompt: `Help me with my chess game. I am playing as ${myColor} and the board is: ${fen}. Please only offer written advice.`
    });
  };


  function onPieceDrop({ sourceSquare, targetSquare }: PieceDropHandlerArgs) {
    if (!gameId || !sourceSquare || !targetSquare || pending) return false;


    const game = gameRef.current;
    if (myColor === "spectator" || game.turn() !== myColor) return false;


    const piece = game.get(sourceSquare as Square);
    if (!piece || piece.color !== myColor) return false;


    const prevFen = game.fen();


    try {
      const local = game.move({ from: sourceSquare, to: targetSquare, promotion: "q" });
      if (!local) return false;
    } catch {
      return false;
    }


    const nextFen = game.fen();
    setFen(nextFen);
    setPending(true);


    stub
      .move({ from: sourceSquare, to: targetSquare, promotion: "q" }, prevFen)
      .then((r) => {
        if (!r.ok) {
          game.load(r.fen);
          setFen(r.fen);
        }
      })
      .finally(() => setPending(false));


    return true;
  }


  return (
    <div style={{ padding: "20px", background: "#f8fafc", minHeight: "100vh" }}>
      {!gameId ? (
        <div style={{ maxWidth: "420px", margin: "0 auto", background: "#fff", borderRadius: "16px", padding: "24px" }}>
          <h1>Ready to play?</h1>
          <p>Start a new match or join an existing game.</p>
          <button onClick={handleStartNewGame} style={{ padding: "12px", background: "#2563eb", color: "#fff", border: "none", borderRadius: "8px", cursor: "pointer", width: "100%" }}>
            Start a new game
          </button>
          <div style={{ marginTop: "16px" }}>
            <input
              placeholder="Paste a game ID"
              value={gameIdInput}
              onChange={(e) => setGameIdInput(e.target.value)}
              style={{ width: "100%", padding: "10px", borderRadius: "8px", border: "1px solid #ccc" }}
            />
            <button onClick={handleJoinGame} style={{ marginTop: "8px", padding: "10px", background: "#0f172a", color: "#fff", border: "none", borderRadius: "8px", cursor: "pointer", width: "100%" }}>
              Join
            </button>
            {menuError && <p style={{ color: "red", fontSize: "0.85rem" }}>{menuError}</p>}
          </div>
        </div>
      ) : (
        <div style={{ maxWidth: "600px", margin: "0 auto" }}>
          <div style={{ background: "#fff", padding: "16px", borderRadius: "16px", marginBottom: "16px" }}>
            <h2>Game {gameId}</h2>
            <p>Status: {serverState?.status}</p>
            <button onClick={handleHelpClick} style={{ padding: "10px", background: "#2563eb", color: "#fff", border: "none", borderRadius: "8px", cursor: "pointer" }}>
              Ask for help
            </button>
          </div>
          <div style={{ background: "#fff", padding: "16px", borderRadius: "16px" }}>
            <Chessboard
              position={fen}
              onPieceDrop={onPieceDrop}
              boardOrientation={myColor === "b" ? "black" : "white"}
            />
          </div>
        </div>
      )}
    </div>
  );
}


const root = createRoot(document.getElementById("root")!);
root.render(<App />);

7. Build and deploy

  1. Build your React UI:
Terminal window
npm run build

This compiles your React app into a single HTML file in the dist directory.

  1. Deploy to Cloudflare:
Terminal window
npx wrangler deploy

After deployment, you will see your app URL:

https://my-chess-app.YOUR_SUBDOMAIN.workers.dev

8. Connect to ChatGPT

Now connect your deployed app to ChatGPT:

  1. Open ChatGPT ↗.
  2. Go to Settings > Apps & Connectors > Create
  3. Give your app a name, and optionally a description and icon.
  4. Enter your MCP endpoint: https://my-chess-app.YOUR_SUBDOMAIN.workers.dev/mcp.
  5. Select "No authentication".
  6. Select "Create".

9. Play chess in ChatGPT

Try it out:

  1. In your ChatGPT conversation, type: "Let's play chess".
  2. ChatGPT will call the playChess tool and render your interactive chess widget.
  3. Select "Start a new game" to create a game.
  4. Share the game ID with a friend who can join via their own ChatGPT conversation.
  5. Make moves by dragging pieces on the board.
  6. Select "Ask for help" to get strategic advice from ChatGPT

Key concepts

MCP Server

The Model Context Protocol (MCP) server defines tools and resources that ChatGPT can access:

TypeScript
const server = new McpServer({ name: "Chess", version: "v1.0.0" });


// Register a UI resource that ChatGPT can render
server.registerResource(
  "chess",
  "ui://widget/index.html",
  {},
  async (_uri, extra) => {
    return {
      contents: [
        {
          uri: "ui://widget/index.html",
          mimeType: "text/html+skybridge",
          text: await getWidgetHtml(extra.requestInfo?.headers.host as string)
        }
      ]
    };
  }
);


// Register a tool that ChatGPT can call to render the UI
server.registerTool(
  "playChess",
  {
    title: "Renders a chess game menu, ready to start or join a game.",
    annotations: { readOnlyHint: true },
    _meta: {
      "openai/outputTemplate": "ui://widget/index.html",
      "openai/toolInvocation/invoking": "Opening chess widget",
      "openai/toolInvocation/invoked": "Chess widget opened"
    }
  },
  async (_, _extra) => {
    return {
      content: [{ type: "text", text: "Successfully rendered chess game menu" }]
    };
  }
);

Game Engine with Agents

The ChessGame class extends Agent to create a stateful game engine:

export class ChessGame extends Agent<Env, State> {
  initialState: State = {
    board: new Chess().fen(),
    players: {},
    status: "waiting"
  };


  game = new Chess();


  constructor(
    ctx: DurableObjectState,
    public env: Env
  ) {
    super(ctx, env);
    this.game.load(this.state.board);
  }

Each game gets its own Agent instance, enabling:

  • Isolated state per game
  • Real-time synchronization across players
  • Persistent storage that survives worker restarts

Callable methods

Use the @callable() decorator to expose methods that clients can invoke:

TypeScript
@callable()
join(params: { playerId: string; preferred?: Color | "any" }) {
  const { playerId, preferred = "any" } = params;
  const { connection } = getCurrentAgent();
  if (!connection) throw new Error("Not connected");


  connection.setState({ playerId });
  const s = this.state;


  // Already seated? Return seat
  const already = this.colorOf(playerId);
  if (already) {
    return { ok: true, role: already as Color, state: s };
  }


  // Choose a seat
  const free: Color[] = (["w", "b"] as const).filter((c) => !s.players[c]);
  if (free.length === 0) {
    return { ok: true, role: "spectator" as const, state: s };
  }


  let seat: Color = free[0];
  if (preferred === "w" && free.includes("w")) seat = "w";
  if (preferred === "b" && free.includes("b")) seat = "b";


  s.players[seat] = playerId;
  s.status = s.players.w && s.players.b ? "active" : "waiting";
  this.setState(s);
  return { ok: true, role: seat, state: s };
}

React integration

The useAgent hook connects your React app to the Durable Object:

const { stub } = useAgent<ServerState>({
  host,
  name: gameId ?? "__lobby__",
  agent: "chess",
  onStateUpdate: (s) => {
    gameRef.current.load(s.board);
    setFen(s.board);
    setServerState(s);
  }
});

Call methods on the agent:

const res = await stub.join({ playerId, preferred: "any" });
await stub.move({ from: "e2", to: "e4" });

Bidirectional communication

Your app can send messages to ChatGPT:

TypeScript
const handleHelpClick = () => {
  window.openai?.sendFollowUpMessage?.({
    prompt: `Help me with my chess game. I am playing as ${myColor} and the board is: ${fen}. Please only offer written advice as there are no tools for you to use.`
  });
};

This creates a new message in the ChatGPT conversation with context about the current game state.

Next steps

Now that you have a working ChatGPT App, you can:

  • Add more tools: Expose additional capabilities and UIs through MCP tools and resources.
  • Enhance the UI: Build more sophisticated interfaces with React.