How to connect Amazon Ads to an AI assistant

The ad MCP category grew up around Meta and Google, because those are the APIs everyone integrates first. Amazon is consistently the gap, which is strange, because for a lot of ecommerce brands it is where a serious share of the budget sits. So Amazon stays a separate tab, a separate weekly export and a separate argument about what ACoS really was.

Amazon is also the platform where the work is most repetitive and most mechanical: harvesting search terms, adding negatives, nudging bids on a few hundred keywords. That is precisely the work worth describing in a sentence. Muze puts Amazon in the same connection as Meta, Google and Shopify, and 36 of its 48 Amazon tools are write operations.

Connect Amazon Ads to your AI assistant in six steps

1. Create a Muze account

   Sign up for Muze. The free tier includes 25 read-only tool calls per month across all connected platforms, with no credit card required.

2. Authorize Amazon Ads over OAuth

   Connect Amazon Ads from your Muze account using your normal Amazon Ads login. You are not filing an Amazon Advertising API application, minting tokens, or handing over a password.

   This is the step that stops most people building this themselves. Amazon Advertising API access normally means an application, a review and a wait, and that is before you write a line of code against a genuinely fiddly API. Because the API approval work sits on the Muze side, your part is a sign-in screen and a permission grant that you can revoke whenever you want.

3. Sync and pick your advertising profile

   Amazon organizes advertising by profile, roughly one per marketplace and account type. Run sync_amazon_profiles once and then list_amazon_profiles to see what you have, so every later request targets the right marketplace.

   Getting this wrong is the classic Amazon reporting mistake: a healthy-looking ACoS that turns out to be a different marketplace from the one you were worried about. If you sell in several regions, note which profile ID maps to which storefront now, while you are looking at the list, rather than rediscovering it in three weeks.

4. Add the Muze connector to your AI client

   Add Muze as a custom connector using the Muze MCP URL: https://backend.muzecmo.com/mcp. The same connector works in ChatGPT, Claude, Cursor, Claude Code, Codex and any other MCP-compatible client.

5. Pull performance and search terms

   Start read-only. list_amazon_campaigns shows the account, get_amazon_performance returns spend and sales, and list_amazon_search_terms surfaces the queries you are actually buying. Ask for a report to be submitted first if the data needs refreshing.

   Spend the first session comparing what comes back against Amazon's own reporting, particularly the date range and the attribution period, because those are where an AI answer and a dashboard usually diverge. Once the reads agree with the console, you can trust the same numbers when they show up inside a cross-platform question later.

6. Act, with a preview in front of you

   On a paid plan you can add negative keywords, adjust keyword, ad group and product target bids, create auto or manual campaigns, and archive what is no longer earning. Each write shows the exact change first and waits for your confirmation, and new campaigns are created paused.

What Amazon data can and cannot tell you here

Be clear about scope before you build reporting on top of it. The Amazon figures available through the connection are advertising figures: ad-attributed spend and sales. They are not your whole Amazon business, and Amazon marketplace revenue is not folded into the Shopify revenue used in blended reporting.

That matters most when you compare channels. If a third of your revenue arrives through Amazon and your blended view only counts store revenue, blended return will look worse than reality. Say the caveat out loud in your own reporting rather than discovering it in a board meeting. The same discipline is covered in the blended ROAS guide, including why platform-reported revenue never adds up to your bank account.

Amazon plus everything else

The reason to run Amazon through the same connection as Meta and Google is that the useful questions are comparative. Where is the marginal dollar best spent this week. Whether Amazon is cannibalizing your own store or growing total demand. Whether your Amazon ACoS is quietly worse than the CPA you accept elsewhere.

None of those can be answered from inside one channel, which is why single-platform tools stay a trap for anyone doing real volume. One connection, 149 tools, and one session that can read all four platforms is the difference between comparing and reconciling.

Prompts worth trying first

• “Which Amazon search terms spent money this month without converting?”
• “Show me my Amazon campaigns sorted by ACoS, worst first, for the last 30 days.”
• “Add the ten worst non-converting search terms as negative exact keywords, but show me the list first.”
• “Compare my Amazon ACoS against what Meta and Google returned last month.”

Frequently asked questions

Is there an MCP server for Amazon Ads?

Yes. Muze covers Amazon Ads with 48 tools, 12 read and 36 write, in the same connection as Meta Ads, Google Ads and Shopify. Most ad MCP servers stop at Meta and Google, so Amazon is one of the harder platforms to reach from an AI assistant.

Do I need Amazon Advertising API access of my own?

No. You authorize over OAuth with your normal Amazon Ads login. The API approval work is already done on the Muze side, so there is no developer application to file and no tokens to manage.

Can AI manage Sponsored Products campaigns?

Yes. Muze can create auto and manual campaigns and ad groups, add keywords and product targets, set and adjust bids, add negative keywords and negative product targets, pull search term reports, and archive campaigns. Every write previews first and requires confirmation.

Can it adjust bids on a schedule?

It can propose and apply bid changes across keywords, ad groups and product targets, including on a recurring basis. Changes still preview before they execute, so scheduled never means unattended.

Which marketplace does it read?

Whichever advertising profile you point it at. Amazon organizes advertising by profile, so sync your profiles first and confirm which profile ID maps to which marketplace before you rely on the numbers.

Does Amazon revenue show up in blended reporting?

Amazon ad-attributed spend and sales are available, but Amazon marketplace sales are not included in the Shopify revenue figure used for blended performance. Treat blended numbers as store revenue against total ad spend, and state that caveat wherever you report it.

What does it cost?

The free tier includes 25 read-only tool calls per month across all connected platforms. Write actions require a paid plan, and Muze never takes a percentage of ad spend.

Keep going