Amazon Ads MCP vs Nova MCP - What Each One Sees 2026

Both servers speak the same protocol and connect to the same assistant, which makes them look interchangeable. They are not. One changes your campaigns, the other tells you whether the change makes money. Here is exactly what each one sees.

What Amazon's Ads MCP server sees

Amazon's Ads MCP server exposes the advertising account through the Model Context Protocol. An assistant can pull campaign performance, create campaigns, and adjust targeting and bids without anyone writing API code. If your question is "set up a new exact-match campaign for these keywords", this is the right tool and it is genuinely good at it.

Its boundary is the ad account. It reports ad spend and ad-attributed sales. It does not know what a unit costs you, what Amazon charged you in referral and fulfilment fees, how many units came back as returns, or how much stock is left. Those numbers live in Seller Central and in your own cost setup, not in the ads API.

What Nova MCP sees

Nova MCP exposes a modeled seller dataset, read-only, across 21 marketplaces with hourly refresh. Advertising is one surface among several: Sponsored Products at campaign, ad group, targeting, keyword and attributed SKU grain, plus the actual customer search queries behind the spend. Around it sit the P&L (40+ Amazon fee types reconciled at SKU level), FBA and AWD inventory with days of cover, current listing health and buy box state, and Best Seller Rank.

The practical difference is what the assistant can conclude. Ads data says a keyword has a 45% ACoS. The joined model says that keyword sits above break-even on a product whose contribution margin is 28%, so the spend is destroying profit even though the campaign looks busy.

The workflow that uses both

Enable both servers in the same assistant and the loop closes:

1. Diagnose with Nova. "Which search terms spent more than $100 last month with zero sales?" comes back as a ranked table in one call, filtered server side.
2. Check the profit side, still with Nova. "For the SKUs behind that spend, what is contribution margin after fees?" tells you whether to cut the term or fix the price.
3. Act with Amazon. Ask the Ads MCP server to add the negative keywords or move the budget.

Nothing in that loop requires a spreadsheet, and nothing in it asks the model to invent a fee calculation. Our PPC analytics feature page covers the same data in the cockpit, and the Amazon MCP hub lists every tool the assistant can call.

Why the split exists at all

Amazon builds against its own surfaces. The ads API is a clean, well-scoped product, so an ads MCP server is a natural extension of it. A profit MCP would require Amazon to model your cost of goods, your inbound freight, your VAT treatment and your multi-account structure, which are things Amazon does not hold.

That is also why community wrappers around the Selling Partner API do not close the gap. They expose endpoints; they do not reconcile them. Amazon documents strict per-operation rate limits in its SP-API usage plans reference, so an assistant paging through raw settlement data hits throttling long before it produces a margin figure.