The short answer: to ask ChatGPT, Claude or Gemini about the sales and profit of your Tiendanube, Shopify or Mercado Libre store, you need an MCP connector. It gives the assistant read access to your real data, with your costs, instead of letting it guess from a pasted file. This isn't about selling inside ChatGPT: it's about asking it about your business and getting an answer built on your numbers.
You ask ChatGPT how much you made last month and it answers with a tidy table, a total and even a comment on your margin. The problem is it made it all up. It has no access to your store: it's completing text that sounds like a report.
Why pasting an export isn't enough
The obvious workaround is to export a CSV and paste it into the chat. It works once, and then it falls apart for three reasons:
- The export has revenue, not profit. It doesn't include product cost, shipping, fees or installments.
- It's stale the next day, and you have to repeat the whole process for every question.
- The assistant doesn't know what each column means, so it confuses revenue with profit and ROAS with real ROAS.
What an MCP is
MCP is the standard AI assistants use to connect to external tools. Instead of pasting data in, you give it a connection: the assistant queries the source when it needs to and answers with what it finds there.
Applied to your store, it's the difference between an assistant that guesses and one that checks. Answers stop being plausible and become verifiable.
What changes when the assistant sees your real numbers
Most of a founder's questions never get asked. Not because they don't matter, but because each one means opening the platform, picking the date range, crossing two reports and exporting to a spreadsheet. Fifteen minutes per question is enough to make you skip it.
When the answer costs one sentence, what you ask changes:
- "How much did I make yesterday and how does it compare to last week?"
- "Which campaigns are losing money this month?"
- "Which product sells a lot but leaves little margin?"
- "Which products are about to run out of stock?"
How Escalafy connects
Escalafy exposes your reports as an MCP server. You set it up once per assistant:
- Open the connector settings in Claude, ChatGPT, Gemini or Grok.
- Paste the Escalafy connector URL.
- Authorize with your account.
- Ask away.
From then on, the assistant reads the same reports you see in the dashboard: same costs, same currency, same time zone. And since the connector also exposes each metric's definition, it doesn't confuse net revenue with profit.
Access is read-only. The connector doesn't touch your store, doesn't touch your campaigns and doesn't move money. It respects the same permissions as the app: if a user has limited reports, financial metrics don't show up in the answer.
What to keep in mind
The last 2 or 3 days keep shifting because of late orders and adjustments from the ad platforms. To compare periods precisely, use closed date ranges. And the quality of the answers depends on your costs being loaded: an assistant connected to incomplete data will hand you the wrong margin, with total confidence.
In short
AI without your data makes things up. With a pasted CSV, it goes stale. With a connector, it answers with your real numbers — and the questions that weren't worth fifteen minutes before start being worth asking.
Keep reading
What Escalafy does for your store
Connect your channels and see the truth behind your numbers. No Excel, in real time.