Once your Amazon data is reachable through an MCP connection, the bottleneck moves from access to asking. These are the prompts that change what a seller does on a Monday, with the phrasing that keeps answers accurate.
Aug 19, 2026 · 9 min · 1,700 words
Once your Amazon data is reachable through an MCP connection, the bottleneck moves from access to asking. These are the prompts that actually change what a seller does on a Monday, grouped by the decision they serve, with a note on why each one is phrased the way it is.
The Model Context Protocol lets an assistant discover the tools a server exposes and call them with typed arguments, as described in the protocol documentation. That means your prompt is really an instruction about which tool to call and with what filter. Three habits do most of the work:
1. Find what actually moved
Compare last 30 days to the previous 30 days. Show the 10 SKUs with the biggest drop in net profit, with net sales, ad spend, units and contribution margin for both periods.
Ranking by profit change rather than sales change filters out the SKUs that only look busy. The paired columns let you see instantly whether the cause is volume, price or spend.
2. Ask why, on one SKU
For the worst SKU in that list, explain the movement: traffic, conversion rate, average price, buy box share, promotions, ad spend and rank, this period versus last.
This is the follow-up that stops the guesswork. A conversion drop with stable traffic is a listing or price problem; a traffic drop with stable conversion is a rank or ad problem.
3. Catch the money-losers
List every SKU with negative net profit over the last 60 days, sorted by total loss, with units sold and the largest fee category for each.
The fee column turns a list into a plan: oversize fulfilment, storage and returns each imply a different fix.
Add one line to every analysis prompt
4. Negative keyword candidates
Show customer search terms from the last 30 days with more than $75 spend and zero orders, ranked by spend, with clicks and CPC.
This is the highest-return prompt in the list. The threshold and the zero-order filter run server side, so you get a short actionable table instead of thousands of rows.
5. Spend above break-even
Which campaigns had an ACoS above the break-even ACoS of the products they advertise last month? Show spend, sales, ACoS and contribution margin.
ACoS alone cannot answer this. It only works when the assistant can reach fee and COGS data alongside ad data.
6. Promote the winners
From auto campaigns in the last 60 days, show search terms with at least 3 orders and an ACoS below 20%, so I can promote them to exact match.
The discovery half of the same report. Most sellers run the waste query and forget this one.
Nova MCP reads your live profit, PPC, inventory and listing data by default. Owner-enabled listing edits are previewed and confirmed before they reach Amazon.
7. What runs out first
List SKUs with fewer than 30 projected days of inventory, sorted ascending, with sales velocity, available units and incoming units.
Projected days of cover, not raw units, is the reorder trigger. Incoming units in the same table stops you double-ordering.
8. Dead stock
Show FBA SKUs with stock on hand and no sales in the last 30 days, with units held and estimated monthly storage cost.
Pairing units with storage cost turns a cleanup task into a ranked one.
9. Silent revenue leaks
Which of my listings are currently suppressed, not buyable or search suppressed? Rank by trailing 90-day sales so I fix the expensive ones first.
Severity alone ranks by Amazon's view. Ranking by your sales ranks by yours.
10. Buy box loss
Show ASINs where I no longer own the buy box, with the current buy box owner, their price and my price.
The price gap column tells you immediately whether it is a hijacker problem or a pricing problem.
The one prompt worth saving as a reusable instruction, because it replaces a dashboard tour:
11. Weekly business briefing
Give me a Monday briefing for last week versus the week before: net sales, net profit and margin by marketplace; the 3 biggest profit gainers and losers by SKU; any SKU with under 21 days of cover; any suppressed or non-buyable listing above $500 in trailing sales; and search terms over $50 spend with zero orders. Keep it under 400 words and end with the three actions you would take first.
It bundles five tool calls into one artefact and forces a prioritised recommendation at the end. The word limit matters, otherwise you get a report nobody reads.
For the connection itself, the Claude setup guide walks through it, and the Nova MCP tools page lists which tool each of these prompts resolves to. Anthropic's original protocol announcement is still the clearest explanation of why this works at all.
Getting reliable answers from an assistant connected to live data
Join thousands of successful sellers who use Nova Analytics to make data-driven decisions and maximize their profits.
Explore more expert insights to grow your Amazon business
Thirteen Amazon MCP servers, ranked by criterion instead of overall. Nova wins profit modelling, SQP, marketplace coverage and multi-account. Amazon Ads MCP wins write actions. The open source repos win on free. Sources and dates on every claim.
Wrapping an API in MCP takes an afternoon. Shipping an MCP a business can run on is everything between your database and the model's first token: context economics, semantics, reconciliation, guardrails and observability.
Teams that connect an assistant to Amazon data discover the cost problem in week two. The fix is rarely a cheaper model. It is changing what comes back from the server, and the same change makes answers more accurate.
