learn by doing
ENESUK
All challenges

Challenge 08

What’s for Dinner?

Build a read-only Mercadona MCP serverAn MCP server publishes a named set of tools that an AI client can discover and call through one connection. and let a Telegram shopping assistant turn any dish into a live, priced product list.

AdvancedComplexity:Time: 60–75 min

Your task

A shopper needs a Telegram assistant: when they send the name of a dish, reply in the same chat with four to six relevant Mercadona products, each current unit priceThe unit price is the current catalog price for one package or sellable unit, not the comparison price per kilogram or litre., and the correctly calculated total for one unit of every product.

Your bonus task

Connect the published catalog MCP serverAn MCP server publishes a named set of tools that an AI client can discover and call through one connection. to an eligible ChatGPT or Claude account, request "paella de marisco" or "gazpacho", and show that the external clientAn external client is another AI application, such as ChatGPT or Claude, connected to the workflow's published MCP address. can use the same three tools.

Example use cases

  • A shopper sends "paella de marisco" and receives a compact list of available Mercadona products with current unit prices.
  • A family names "gazpacho" and gets one practical package of each essential ingredient plus the basket total.
  • A cooking workshop can reuse the same catalog tools from Telegram, ChatGPT, or Claude without giving the model permission to change data.

Before you start

  1. 01

    Sign up for n8n Cloud or use an updated n8n instance with a public HTTPS address; read the MCP Server Trigger guide and use its Production URL after publishing.

  2. 02

    Create a Telegram account, make a bot with BotFather, and add the token by following the n8n Telegram credentials guide.

  3. 03

    Create an OpenRouter account, generate an API key, and store it in an n8n OpenRouter credential.

  4. 04

    Keep the three read-only catalog endpoints ready: all categories, one category example, and one product example. These are live third-party endpoints and may change. Open each link in your browser before you start: if one stops returning JSON, the endpoint has changed, not your workflow.

  5. 05

    For the bonus, confirm that your account supports remote custom connectors and follow either the ChatGPT MCP connector guide or the Claude custom connector guide. The server is intentionally read-only; deactivate the workflow after testing if you leave authentication set to None.

What the workflow must do

  1. 01

    Every text message containing a dish name receives one Telegram reply with four to six relevant, available Mercadona product names and current unit prices, with no internal IDs in the response.

  2. 02

    Execution evidence shows that the agent calls get_categories first, then get_category with its required id for each selected category, and finally get_product for every product it includes.

  3. 03

    The reply ends with a two-decimal euro total calculated for one unit or package of every listed fixed-price product, and two different dish tests use only values returned by the live catalog toolsThese tools read the current category and product records directly from the catalog endpoints while the workflow runs..

Solution

See the workflow challenge answer

Reveal the completed workflow only when you are ready to compare it with your own.

Submit as solved

Ready to submit?

Submit when your team has a working workflow to demonstrate.

Solution

Reveal the workflow solution?

This will show the completed workflow for this challenge. Are you sure you want to continue?

Mentor review

Find a mentor and ask them to review.

Show them your working workflow. Once they approve it, collect the balloon for this challenge.

Start next challenge