# Connect restaurant data to AI with Feast

Feast's restaurant data connector is free. Start at https://feastalytics.com/connect-ai, choose your POS and preferred AI, and create your account. Connect the restaurant's POS in onboarding and book the setup call. Toast, Square and Clover have setup options. For another POS, the team confirms the connection with you.

## Connect a supported AI client

The hosted MCP server is https://mcp.feast-api.com/mcp. Choose an HTTP connection with OAuth in a client that supports remote MCP servers. Sign in to Feast and authorize the connection. Your Feast account must have access to the restaurant. Your AI client's plan or workspace settings may affect connector availability.

### ChatGPT

1. Open [ChatGPT Plugins](https://chatgpt.com/plugins). Choose **Add**, then **Add custom MCP server**.
2. Name it **Feast**. Under Connection, choose **Server URL** and enter `https://mcp.feast-api.com/mcp`.
3. Select **OAuth**. Review the warning, then choose **I understand and want to continue** and **Create as a plugin**.
4. Open the resulting plugin and install it. Connect and sign in to Feast when prompted.
5. Start a new chat, type `@` and select **Feast**. Name your restaurant in the first request.

If the custom-server option is missing, check your plan or workspace's plugin permissions. See [OpenAI's current connection guide](https://developers.openai.com/plugins/deploy/connect-chatgpt) for the platform setup.

The downloadable Feast skill is published with the [latest CLI release](https://github.com/feastalytics/cli/releases/latest). It adds Feast-specific instructions. Feast's in-app guide, "Connect Feast to ChatGPT", includes the skill installation steps and screenshots.

### Claude Code

Add the remote server with:

```sh
claude mcp add --transport http feast https://mcp.feast-api.com/mcp
```

Complete the client's OAuth sign-in. Select the restaurant before asking for its data. If your client does not expose a remote MCP connection, tell the setup team which client you use so they can confirm the available path.

## Start with one useful answer

Try this after the POS data is available:

"For my restaurant, compare the last two complete Monday-through-Sunday weeks. Show the dates, timezone, sales definition, orders and average order value. Tell me what data is missing before drawing conclusions."

Check the totals against the POS for the same restaurant, dates and sales definition. Data still importing is missing data, not zero sales. A change between weeks does not by itself explain the cause.

## Tasks to try with connected data

These are starting requests for your AI. Available history and fields depend on the connected source. They are not preconfigured automations.

### Compare dayparts

"Compare lunch and dinner over the last four complete weeks. Show the time boundaries you used, orders, sales and average order value for each daypart. Keep closed days separate from days with no data."

Use the result to find where there is room for more business. Check capacity and opening hours before interpreting a low total as weak demand.

### Compare locations

"Compare my locations for the same complete date range. Show sales, orders, average order value and which locations have incomplete history."

Confirm location identity, local dates and coverage. A newly connected location should not look like the worst performer because its older data is missing.

### Prepare a weekly sales summary

"Create a weekly summary with one row per date, showing sales, orders and average order value. Include the restaurant, timezone, sales definition and missing dates. Give me the table in a format I can copy into Google Sheets."

This requests a table. An automatically refreshed spreadsheet needs an additional connection and a schedule. Bring that workflow to your setup call.

### Investigate a sales decline

"My sales were lower last week. Compare the same weekdays, then break down the difference by orders and average order value. Show which conclusions the available data supports and which questions need more data."

If you want to examine marketing, weather, new guests or returning guests, first establish whether those data are available. Do not turn a sales comparison into a causal claim.

### Plan a custom workflow

"I want Toast data in Google Sheets and a text alert when a day's sales fall below a threshold. List the data, permissions, destination and schedule this needs. Help me define the threshold before building anything."

Feast can help you work through the setup. A recurring export or text alert requires its own implementation; asking the question does not activate one.

## For agents using Feast tools

Start with `listOrganizations` to establish the authorized restaurant. Use `describeData` to discover the current catalog and query grammar, then `queryData` with the documented fields. Do not invent object names or assume the catalog matches another restaurant's connection.

Use the account's available data and return explicit dates, units, definitions and missing fields. Some Feast tools change account data. A data question needs reads; a request to build or run something needs the owner's intended action and the appropriate permissions.

## Continue with Feast

Start your free setup: https://feastalytics.com/connect-ai

Explore Feast's restaurant revenue analytics and marketing: https://feastalytics.com/