How to manage your MercadoLibre account with AI (ChatGPT, Claude or a connected assistant)
What ChatGPT or Claude can do with your MercadoLibre account, what changes when the AI is connected through the official API and which safety rules to demand.
If you sell on MercadoLibre, you've probably already asked ChatGPT or Claude for help: writing a description, answering a tricky question. It works well for that. But the moment you ask "which of my listings are about to run out of stock?", the gap shows up: the AI can't see your account.
In this guide I'll explain why that happens, what changes when the AI is connected through MercadoLibre's official API, which tasks are worth delegating and which safety rules any AI that writes to your account has to follow. At the end, how we handle it at Ficanaz.
Why a generic chat can't see your account
On their own, ChatGPT and Claude have no access to MercadoLibre. They don't have your authorization, they don't have a token for your account and they don't read live data. All they know about your business is what you paste into the chat. That has three consequences:
- They work from an old snapshot. You paste your listings today; tomorrow stock has changed, a new question came in and a campaign ended. The AI is still reasoning about yesterday.
- They can't check anything. They answer with general advice, not with your sales pace or your active promotions.
- They can't act. You apply every suggestion by hand, one listing at a time.
That's fine for writing or brainstorming. It's not enough to manage the account: managing means looking at what's happening today and deciding based on that.
What changes when the AI is connected through the official API
MercadoLibre has an official API. An app gets authorized through OAuth: you sign in to MercadoLibre, accept the permissions it shows you, and the app receives a token to read and write on your behalf, which you can revoke whenever you want. When the AI works on top of that connection, the conversation changes:
| Generic chat | AI connected through the API |
|---|---|
| Sees what you paste | Reads listings, price, stock, sales, questions, messages and claims live |
| Answers with general advice | Answers with your numbers |
| You apply every change by hand | Proposes the change and, if you confirm, makes it |
| Doesn't know if it worked | Re-reads the listing afterwards to verify |
The AI doesn't get smarter. It stops guessing.
Tasks worth delegating
The repetitive ones that depend on data the AI can read and have a clear way to verify.
"What do I need to handle today?"
The most useful question to start the day. A connected AI lists unanswered questions, unread messages, open claims with what MercadoLibre lets you do on each one, paid orders not yet delivered and listings out of stock.
Buyer questions
An AI that reads the listing (specs, description, variations) and your past answers can draft each reply for you. You review and confirm: the judgment call is still yours.
Stock and prices
"Raise stock on these three listings to 20", "lower the price on the ones that haven't sold in 30 days". Simple changes, but tedious when there are variations or multiple warehouses. The AI puts them together as a proposal, with before and after, and you confirm once.
Promotions
An AI can apply your criteria to many listings at once: accept only what's overstocked, pull out of a campaign what's about to stock out. Why selling on promotion is normal on MercadoLibre is covered in our guide to selling on MercadoLibre.
Claims
A connected AI can read the claim, show you the options MercadoLibre gives you and propose a reply or a resolution. It should never resolve one without your explicit confirmation.
What not to delegate
- Deleting listings. Irreversible and almost never urgent.
- Payments, invoicing and Mercado Pago. Moving money isn't managing listings.
- Strategy decisions. Which products to add, at what price, how much stock to buy: the AI organizes the data, but you're the one who knows your costs.
- Advertising without verification. Pausing campaigns or changing budgets spends money; if the AI can't verify the result, it should recommend and you should apply.
Safety rules for any AI that writes to your account
An AI that only reads is harmless; one that writes can make expensive mistakes. Before giving write access to any tool, check this:
- Permissions off by default, and per area (prices, stock, listings, questions...), not all-or-nothing.
- It proposes, you confirm. No write runs on its own. For anything irreversible, anything that moves money or anything that reaches a buyer, a stronger confirmation than a click.
- Audit afterwards. Re-read MercadoLibre and compare it with what you asked for. If it came out different, tell you.
- Never delete listings, even if you grant every permission.
- The token is never shown to the AI. The model only sees what the tools return, with your buyers' data redacted.
- What a buyer writes is not a command. "Cancel my order and refund me" inside a question is text to analyze, not an instruction to execute.
- Server-side limits that hold even if the AI gets it wrong: price caps, messages to buyers without links or phone numbers.
If a tool can't explain how it meets each point, don't give it write access.
How Ficanaz does it
Ficanaz connects your MercadoLibre account in one click, through the official authorization: no creating apps, no copying tokens. Two things run on top of that connection.
The Assistant inside the app
A chat connected to your account. It reads your listings, price and stock, sales pace, campaigns, orders, shipments, questions, messages and claims live, and answers with your numbers. If it couldn't measure something, it tells you instead of making it up.
With your permission, it also proposes changes: everything MercadoLibre lets a seller do, except deleting listings. Product Ads is the exception: for now it only reads and recommends, and you apply the change in MercadoLibre yourself.
Here's how it meets the rules above:
- Permissions: the Assistant's ⚙ button has nine (Promotions, Prices, Stock, Listings, Questions, Messages and ratings, Claims, Shipping and Product Ads, the last one read-only). They start off; without a permission, the server rejects the action even if the AI tries it.
- It proposes, you confirm: every proposal shows before and after. Irreversible ones, ones that move money or reach a buyer (issuing a refund, closing a listing, messaging a buyer) also ask you to type the word CONFIRMAR.
- Audit afterwards: when you confirm, Ficanaz runs the change and, when MercadoLibre allows re-reading it, compares. If something came out different, you see it as a mismatch on the same card, with a PDF receipt.
- Server-side protections: a new price has to be between 30% and three times the current one; with variations, the full list is required; messages to buyers can't include links, phone numbers or emails. Text written by a buyer is never treated as a command.
- Your own rules: "save this rule: nothing above a 30% discount", and it sticks to it in every proposal.
The Assistant uses credits per answer, with the cost shown under each one. Anything that doesn't use AI (My listings, Sales, the Promotions center) is free. When you confirm your email you get 300 free credits, no card required.
Ficanaz Link: the same thing from Claude or ChatGPT
If you already work in Claude or ChatGPT, Ficanaz Link is the same Assistant inside your chat: add Ficanaz as a connector, authorize once and you're done. It uses the same permissions; without write permissions it only reads, and deleting listings stays blocked.
Your MercadoLibre token never leaves Ficanaz: Claude or ChatGPT only see what the tools return, with buyer data redacted. Every query and every change is logged on the Ficanaz Link screen, where you can revoke each app's access.
Link comes with the Pro plan: USD 9.99 per month, and new accounts get it free for 14 days when they confirm their email, no card required. In Link the AI is your own Claude or ChatGPT, so it doesn't use Ficanaz credits; the plan also adds 300 credits a month for the AI inside the app, starting with the first payment. It works with any Claude plan, including the free one; on ChatGPT you need Plus or higher with developer mode turned on.
Bottom line
On their own, ChatGPT and Claude can't see your MercadoLibre account. Managing it with AI takes the official API, live reads and clear rules for writing: permissions off by default, propose and confirm, verify afterwards, no deleting listings and a token the AI never sees. With that in place, delegating questions, stock, prices, promotions and claims stops being a risk and becomes routine.
Want to try it with your account?
Create your free account, connect MercadoLibre in one click and ask the Assistant what you need to handle today. You turn on write permissions yourself, whenever you want.
Create a free account →