Amazon MCP with Claude: Setup, Prompts & Use Cases

Key Takeaways

1. MCP is an open standard that lets Claude call Amazon tools (Ads, Selling Partner API) using plain-English prompts.

2. Amazon's own Ads MCP server is in open beta and works with Claude, ChatGPT and Gemini, but reports can take 10–30 minutes.

3. Start read-only. Let Claude analyse first, and approve every spend or bid change yourself.

4. Brand Analytics (Search Query Performance, Market Basket) is the best data to pair with Claude for keyword and listing decisions.

5. "Amazon MCP" can also mean the AWS (Amazon Web Services) MCP servers. They are for cloud developers, not sellers.

6. Atom11 adds the retail layer Claude lacks on its own (inventory, buy box, pricing, rank) in one dashboard, with Neo diagnosing sales changes across 15+ signals.

Amazon Ads brought in $17.7 billion in Q3 2025, up 24% year on year. More ad spend means more reports to read, and most sellers still lose hours to exports and pivot tables. Amazon MCP with Claude shortens that. You connect Claude to your Amazon data once, then ask questions like "Which campaigns wasted spend last week?" and get an answer in plain English.

This guide shows what Amazon MCP is, how to connect it, which prompts work, and where it falls short.

What does "Amazon MCP" mean?

MCP stands for Model Context Protocol. It is an open standard, introduced by Anthropic in late 2024, that lets an AI assistant such as Claude use outside tools and data. Think of it as a universal plug. Claude is the device, and your Amazon account is the socket.

"Amazon MCP" currently refers to three different things:

If you sell on Amazon, the SP-API route is the one that matters most, because it exposes your orders, inventory and listings. Our seller MCP guide explains how it works and how sellers can use it day to day.

How Amazon MCP works with Claude

1. You write a prompt in plain English.

2. Claude picks a tool exposed by the MCP server, such as "create campaign" or "get report".

3. The server calls Amazon's API using credentials you approved through OAuth.

4. Claude reads the result and explains it, or builds a table.

Amazon's Ads VP of Ads Measurement, Paula Despins, described the server as a "translation layer" that turns conversation into structured API calls. MediaPost reports that connected agents can manage campaigns, check reports, change account settings and view billing data.

Important limits from the beta: the server supports only MCP "Tools" for now, reports are asynchronous and can take 10 to 30 minutes, and ad data is kept for 60–95 days depending on ad format. Plan your analysis around that window.

How to connect Claude to Amazon: 5 steps

Setup takes about 10 minutes for managed connectors, according to Seller Labs' setup guide (a managed-connector vendor). Self-hosting takes longer.

1. Check your access. Amazon's Ads server is available to Amazon Ads partners with active API credentials. Most sellers connect through a partner tool instead.

2. Pick a route. Managed (hosted connector, fastest) or self-hosted (open-source repo, needs Python and SP-API keys).

3. Authorize Amazon. Approve permissions through the standard Selling Partner or Ads OAuth screen.

4. Add the connector to Claude. In Claude, open Settings → Connectors → Add custom connector, paste the server URL and sign in. For Claude Desktop config files, fully restart the app afterwards. Porter Metrics' public tutorial shows the same flow for Amazon Seller data.

5. Start with a read-only key. Test with questions only. Add write access later, and keep human approval on anything that spends money.

Common Amazon MCP Setup Errors and How to Fix Them

Even a clean setup can hit small snags. The table below lists the four problems sellers run into most often, what usually causes each one, and the quickest fix so you can get Claude reading your Amazon data again.

Here is the same information in detail:

• Connector not showing. This usually happens because the app was not restarted after you added the connector. Quit Claude fully and reopen it.

• JSON error. This is usually caused by a missing comma in the config file. Paste the file into a JSON checker to find and fix the mistake.

• Empty data. This usually means there is no active plan on the account or the wrong marketplace is selected. Confirm the account and region you connected.

• Slow first answer. The connection is often still warming up. Retry the prompt, and later prompts are usually faster.

12 Claude prompts for Amazon sellers

Specific prompts beat vague ones. Always name the date range, marketplace and ASIN.

Advertising

1. "List campaigns from 1–30 September with ACoS above 35% and spend above $100. Sort by wasted spend."

2. "Show search terms with 15+ clicks and zero orders. Suggest negative keywords by match type."

3. "Compare Sponsored Products ROAS this week vs. last week. Explain the biggest changes."

4. "Draft a new Sponsored Products campaign for ASIN [B0XXXXXXX] with a $50 daily budget. Do not launch it. Show me the plan."

Inventory and listings 

5. "Which ASINs have under 14 days of stock but active ads?" 

6. "Show ASINs where I lost the Buy Box in the last 7 days and whether ads still ran." 

7. "List listings with a conversion rate under 8% and more than 200 sessions."

Brand Analytics 

8. "From Search Query Performance, list queries where my impression share is under 5% but click share is above 15%." 

9. "Using Top Search Terms, find ten terms my competitors win that I do not bid on." 

10. "From Market Basket, which products are bought with my best seller? Suggest bundle ideas."

Reporting 

11. "Build a weekly summary table: spend, sales, ACoS, TACoS, top 3 wins, top 3 problems." 12. "Explain in plain English why sales fell 20% last week. List what you checked and what you could not check."

Tip: Add "Tell me if the data is incomplete" to every prompt. It reduces confident but wrong answers.

Using Amazon Brand Analytics with Claude

Brand Analytics is the richest free data source for sellers. Amazon's seller blog lists seven dashboards, including Search Query Performance, Top Search Terms, Repeat Purchase Behavior and Market Basket Analysis. You need to be enrolled in Brand Registry. Data usually goes back three years and appears within about three days.

Why pair it with Claude? The raw reports are huge. Claude can scan thousands of rows and flag the keywords worth bidding on, or the ASINs that hurt your click share. Combine this with ad data and you can answer a question few sellers can: "Which high-volume search terms am I not funding at all?"

Where Amazon MCP falls short

• Slow reporting. The 10–30 minute report time makes it poor for real-time bidding.

• Silos. The Ads server sees ads only. The SP-API sees listings and inventory. Claude cannot see costs such as freight or duties unless you supply them.

• Fragmented attribution. As Clear Ads Agency notes, MCP "does not fix Amazon's fragmented attribution model."

• Security. Third-party MCP servers vary in quality. Use read-only keys, review what each tool can do, and never paste credentials into a prompt.

• Not an optimizer. Claude answers questions well. It does not run 24/7 bid rules, dayparting or inventory-based pausing for you.

Why Claude + MCP Works Best with Atom11

Claude with MCP is a strong question-and-answer layer. But running Amazon ads every day needs something that is always on, knows your stock and prices, and lets you undo a bad change. That is the gap Atom11 was built to fill.

Atom11 is a retail-aware Amazon PPC platform, built by ex-Amazonians for brands, vendors and agencies. It links your ads data with inventory, pricing, buy box and organic rank, then shows it all in one performance dashboard. Here is what that adds on top of Claude and MCP:

• Faster root-cause answers. Claude can answer one question at a time. Atom11’s AI copilot, Neo, compares more than 15 ad and retail signals at once to explain why sales moved, including causes outside your ads.

• Inventory-aware bidding. Instead of asking Claude to pause ads on low-stock products, Atom11 adjusts bids automatically using inventory rules, and re-enables ads when stock returns.

• One dashboard, not a rebuild each time. Ads, retail and pricing data sit in one view, with links to Power BI, Google Sheets and Looker for custom reports.

• Automation you can check. The performance monitor shows whether an automated change helped, and version control lets you roll back, for example around Prime Day or BFCM.

• Market visibility. Digital shelf tracking helps you watch competitors and ad suppressions, which MCP connectors do not cover.

Who gets the most value:

• Brands and sellers scaling spend without adding headcount.

• Agencies running many accounts that need one dashboard and shared workflows.

• Large-catalog brands managing thousands of ASINs.

A practical setup many teams use: let Claude handle exploratory questions and Brand Analytics research, and let Atom11 run the automation, dashboards and root-cause analysis. If you want to see it on your own data, use the Contact Us option on the Atom11 website to book a walkthrough.

Conclusion

Amazon MCP with Claude turns hours of report-wrangling into a conversation. Connect with read-only access, use specific prompts, pair ad data with Brand Analytics, and keep a human in charge of spend. Be realistic about beta limits: slow reports, short data retention and siloed data. For daily automation, a single dashboard and fast root-cause answers, add a retail-aware platform such as Atom11. If you want to see how that looks on your own accounts, Book a demo with us today!.