This repository contains a NextJS starter app built on top of the Responses API. It leverages built-in tools (web search and file search) and implements a chat interface with multi-turn conversation handling.

Features:

• Multi-turn conversation handling
• Streaming responses & tool calls
• Function calling
• Display annotations
• Web search tool configuration
• Vector store creation & file upload for use with the file search tool
• MCP server configuration
• Google Calendar & Gmail integration via first-party connector

This app is meant to be used as a starting point to build a conversational assistant that you can customize to your needs.

How to use

1. Set up the OpenAI API:

   • If you're new to the OpenAI API, sign up for an account.
   • Follow the Quickstart to retrieve your API key.

2. Set the OpenAI API key:

   2 options:

   • Set the OPENAI_API_KEY environment variable globally in your system
   • Set the OPENAI_API_KEY environment variable in the project: Create a .env file at the root of the project and add the following line (see .env.example for reference):

   OPENAI_API_KEY=<your_api_key>

3. Clone the Repository:

   git clone https://github.com/openai/openai-responses-starter-app.git

4. Install dependencies:

   Run in the project root:

5. Run the app:

   The app will be available at http://localhost:3000.

Tools

This starter app shows how to use built-in tools, MCP servers, and first-party connectors with the Responses API.

You can configure these tools directly from the UI, but some tools require additional setup (e.g. Google OAuth).

Built-in tools

We have several out-of-the-box tools available to use with the Responses API. This demo app implements and allows to configure directly from the UI the following tools:

• File search, to allow the model to access your files in a vector store
• Web search, to allow the model to search the web
• Code interpreter, to allow the model to run Python code to solve problems

Other built-in tools, such as computer use or image generation, are not implemented in this demo app.

MCP servers

The UI allows you to configure a public MCP server to use with the Responses API. If you want to use an MCP server that requires authentication, feel free to update lib/tools/tools.ts to add your own logic. You can use the Google connector integration as an example of how to use access tokens.

Custom functions

This demo app comes with example functions, get_weather and get_joke. You can add your own functions to the config/functions.ts file.

Google integration

This app shows how you can use OpenAI's 1P connectors to integrate with Google and let the assistant read your calendar and email inbox. The app performs a secure OAuth flow in your browser, stores tokens per session, and attaches the Google connector to the Responses API tools list with your access token.

To test this instructions, read the instructions below to set up the Google OAuth 2.0 client and enable the Google Calendar and Gmail APIs.

Learn more about the available 1P connectors in our documentation.

Setup (Google OAuth)

1. Create an OAuth 2.0 client for a Web application in your Google Cloud project (see documentation for accessing Google APIs with Oauth 2.0 docs).

   • In Google Cloud, go to APIs & Services > Google Auth platform > Clients > Create client > Web.
   • Add your redirect URI: http://localhost:3000/api/google/callback.
   • Copy the client ID. Create and copy a client secret.

2. Enable APIs in the same project:

   • Google Calendar API
   • Gmail API

3. Configure data access scopes in Google Auth Platform to match what you need. This demo uses:

   • openid
   • email
   • profile
   • https://www.googleapis.com/auth/calendar.events
   • https://www.googleapis.com/auth/gmail.modify

4. Create .env.local (you can copy .env.example) at the project root and add:

   GOOGLE_CLIENT_ID="your-google-client-id"
   GOOGLE_CLIENT_SECRET="your-google-client-secret"
   GOOGLE_REDIRECT_URI="http://localhost:3000/api/google/callback"

Demo flows

Try web search + code interpreter

After enabling web search and code interpreter in the UI, ask the model:

"Can you fetch the temperatures in SF for August and then generate a chart plotting them?"

The model should use the web search tool to fetch the temperatures and then use the code interpreter tool to generate a chart which will be displayed in the UI.

Try file search

• Save PDF files, for examples blog posts (you can use this one, then print the page and use the "Save as PDF" option)
• Create a new vector store and upload the PDF file(s)
• Enable file search and ask the model a question which can be answered by the PDF file(s), for example:

  "What's new with the Responses API?"

• The model should use the file search tool to find the relevant information in the PDF file(s) and then display the response

Try the Google integration

• Click "Connect Google Integration" in the UI and complete the OAuth flow; you will be redirected back with connected=1.
• Ask the assistant to perform tasks—for example, "Show my next five calendar events," or, "Summarize the most recent wirecutter emails".
• The app will attach Google Calendar and Gmail connectors (via MCP) to the tools list using your access token and stream results back to the UI.
• To invalidate the OAuth session, clear the app cookies (Chrome DevTools > Application > Storage > Cookies). If you only clear gc_access_token, the app will use the gc_refresh_token to refresh without re-authenticating.

Contributing

You are welcome to open issues or submit PRs to improve this app, however, please note that we may not review all suggestions.

License

This project is licensed under the MIT License. See the LICENSE file for details.