AI Assistant Access

Ask your assistant, not your spreadsheet.

MarketplaceHub speaks the Model Context Protocol (MCP), so Claude or ChatGPT can read your account directly. Ask how many units of a SKU are left, which connection is failing, or whether last night's price push finished. The answer comes from your own data.

  • No API key to copy or paste
  • Twenty-three tools: seventeen read, six write
  • You approve it on a consent screen
  • Every call in your request log
Twenty-three tools

Seventeen tools that read. Six that write.

It is the same data the REST API serves, through the same permissions and the same request allowance. Nothing new is exposed to an assistant that is not already yours.

Ask

Read

Ask about your catalogue

Products and their listings, one SKU at a time or paged through the lot. Plus marketplace categories, the attributes each one requires and the values it accepts.

Ask what's broken

Which marketplaces are connected and the health of each, including the one that needs reconnecting. Which background imports, publishes or syncs are waiting, running or failed, and whether one really finished.

Act

Write · needs your approval

Let it set price and stock

Name the SKUs and the new figures, and the change is pushed out to your connected marketplaces. Only if you granted the write permission for inventory.

Let it confirm a shipment

Carrier, service and tracking number against the order items you name. The same confirmation the Orders queue sends, from a conversation.

The other four write tools
  • Ask Amazon to ship an order from your FBA stock, after showing you its fee
  • Cancel that shipment
  • Buy a shipping label through Amazon, after showing you its price
  • Cancel that label

Never twice. All six write tools are keyed. A repeat of the same call returns the first answer instead of doing the work again.

Guardrails

What it deliberately cannot do

An assistant is new. The first things you hand one should not be the expensive ones.

Publish a new listing

Publishing writes to a marketplace against a schema that has to be exactly right. That is a poor first thing to hand an assistant, so it is not offered at all.

Not offered · Publishing a listing

Refund a buyer

Money moves, and it does not move back on its own. Refunds stay with a person, in the Orders queue or through your own code on the API.

Stays with a person · Refunds

See buyers by accident

A buyer's name and address are left out of every order answer, unless the assistant asks for them on purpose. Most questions about an order do not need them.

Left out · Buyer name and address
Connecting

You stay in charge of it

Approving is what creates the access. Nothing happens before you do, and nothing is hidden afterwards.

  1. 1

    Add one address

    In any client that supports a remote MCP server with OAuth, add https://api.marketplacehub.com/mcp as a custom connector. The assistant registers itself.

  2. 2

    Sign in and approve

    You finish on a normal MarketplaceHub sign-in page and see exactly what was asked for. That approval is the access.

  3. 3

    Ask, and read it back

    Every call the assistant makes lands in your own request log, so you can see what it looked at and when.

A consent screen, per assistant

After you sign in you are shown exactly what was asked for. An assistant can never be granted more than your account offers, and connecting one connects nothing else.

Every call is logged

An assistant's calls appear in your request log next to your REST traffic. It draws on the same allowance, not a second one.

Disconnect in one click

Every assistant you approved is listed in your account, with what it may do and who approved it. The list belongs to the account, so a colleague can disconnect one too.

Read the MCP docs
Agent Bot

Or ask it without leaving MarketplaceHub

The Agent Bot is built into the Merchant app, so there is nothing to connect.

Each answer uses 1 AI credit. A question that gets no answer costs nothing.

Open it where you work

Press the Agent Bot row in the sidebar, or the robot button at the top of the screen on a phone. No setup, and no other assistant to connect.

Ask about your account

It reads your products, orders, marketplaces, background jobs and AI credits, and links you to the right page for what it found.

Only what your role allows

Each team member sees only what their own role lets them see, by the same rule the pages themselves follow.

It proposes, you decide

It can propose a price or stock change, a shipment confirmation, an AI re-run, or one missing value filled in on the listings that lack it. Nothing changes until you press Confirm on a card drawn from your own data, not from what the assistant says.

  • Confirming a re-run only opens the AI Launchpad; nothing is billed until you press Launch there.
  • Filling a value sends nothing to a marketplace.

Questions about assistant access

Which assistants, what they can reach, and how to stop one.

Still wondering?

Our team replies within a few hours, Mon–Fri.

Which assistants can connect?

Any client that supports a remote MCP server with OAuth. You add https://api.marketplacehub.com/mcp as a custom connector, the assistant registers itself, and you finish on a normal MarketplaceHub sign-in page. Older command-line tools may not support remote servers yet.

Do I have to create an API key for this?

No, and that is the main difference from the REST API. There is no secret to copy and nothing to paste back. You sign in, approve what the assistant asked for, and that approval is the access. If you'd rather write code against the API instead, that's the Developers route.

Can it change my prices or ship my orders?

Only if you approved the matching write permission. Six of the twenty-three tools write: one sets price and stock for SKUs you name, one confirms shipment with carrier and tracking, one asks Amazon to ship an order from your FBA stock after showing you its fee, one cancels that shipment, one buys a shipping label through Amazon after showing you its price, and one cancels that label. An assistant approved for reading is never even offered them, and the consent screen names them before you agree.

What if the assistant retries and does something twice?

It can't. All six write tools are keyed, so a repeat of the same call returns the first answer instead of doing the work again. A lost reply cannot confirm one shipment twice or apply the same price change twice. Nothing is needed at your end; it is on by default.

What is "Stay connected" on the consent screen?

It is not a permission over your data. It lets the assistant keep working without sending you back to a sign-in page every hour. It reads nothing extra, and disconnecting the assistant ends it along with everything else.

How do I stop an assistant?

Open API → Assistants in your account and press Disconnect. It then has to ask for your approval again before it can read anything. A token it already holds keeps working until it expires, up to an hour, so contact support as well if you think one has leaked. To stop every assistant at once, use Pause on the Approvals tab. How approvals, limits and the change record work: AI agents and your store.

Give your assistant your own data, on your terms

Connect Amazon, eBay, Shopify and Etsy in one place.

  • No hidden fees
  • Cancel anytime
  • Setup in 4 minutes