Live now
Nova MCP for Claude is live
Talk to your full Amazon dataset (P&L, fees, PPC, inventory) from Claude. Free for users signing up before 07/31. Setup takes under 2 minutes.
The Amazon Ads MCP Server opened a door. The bigger room is on the other side of it: AI agents that read the entire seller business, not just the ad account. That is the integration Nova is shipping over the coming weeks.
The interesting question is not "can AI manage my ads"
When the Amazon Ads MCP Server went into open beta in March 2026, the reaction focused on speed: any AI client could now spin up campaigns in seconds. That is real, and it changes what a PPC agency does day-to-day. But it sidesteps the harder question.
The harder question is whether an AI agent can reason about your business, not just your ad account. Can it tell you which 10 SKUs lost contribution margin last week after returns and fees? Can it explain why TACoS rose in Germany but fell in the US? Can it draft a saved view that breaks down P&L by category and season? Those questions need data the Ads MCP Server cannot see , margins, fees, inventory, COGS, organic.
The answer to whether an AI can do that depends almost entirely on one thing: the data model it gets to read.
MCP standard announced
Anthropic published the Model Context Protocol , an open standard that lets AI clients connect to external tools and data sources through a uniform interface.
Amazon Ads MCP Server open beta
Amazon exposed the ad layer to MCP-compatible AI clients. Campaign creation in seconds, but visibility limited to the ads silo.
Nova ships Claude + MCP integration
Nova exposes the full seller data model , P&L, fees, PPC, FBA inventory, organic across 23 marketplaces , to Claude over MCP. Private waitlist open.
What Nova is shipping
Over the coming weeks Nova is rolling out a Claude integration over the Model Context Protocol. Once a Nova workspace is connected, Claude can read the full Nova dataset , orders, every Amazon-reported fee, COGS, product-level PPC spend, FBA inventory, organic, across 23 marketplaces, refreshed every 15 minutes , and answer questions about it in plain English.
The integration reads by default. Listing edits stay off until Nova turns them on for your workspace, and every edit is previewed and confirmed before it reaches Amazon. Prices and stock levels are never changed. It reads what is already in your Nova analytics and lets Claude reason about it. Three concrete capabilities at launch:
- 1.Ask in plain English. "Which SKUs dropped the most contribution margin in the last 30 days?" Claude reads it from the Nova data model and answers with the numbers, citing the underlying fields.
- 2.Build views from a prompt. Describe a breakdown (by category, brand, season, marketplace) and Claude drafts a custom breakdown in Nova. You review and save it.
- 3.Summarize period-over-period. "This week vs last week across revenue, ad spend, fees and refunds" , get a written summary backed by the actual rows, not a guess.
Why the data model is the moat, not the model
Anthropic, OpenAI and Google ship a better large language model every few months. The frontier moves fast. But none of those models can answer "did this SKU make money last week" unless something exposes the joined-up data behind the question. That something is the data model.
The Nova data model was built for exactly this work over the last few years. It normalizes 900+ metrics across 23 marketplaces, attributes 40+ Amazon fee types at the SKU level, joins product-level PPC spend to revenue and margin, and brings FBA inventory into the same surface, refreshed every 15 minutes. That joined-up view is what lets an AI agent answer profit questions instead of revenue questions.
A simple test
Ask any "AI for Amazon" tool: "Which 5 SKUs lost the most contribution margin last week after fees, refunds and ad spend, in which marketplaces?"
Most tools either cannot answer it (they only see one slice), or they answer with revenue numbers dressed up as profit. The answer that holds up is the one where the AI is reading a data model that actually joins those fields. That is the bar.
How this fits with the Amazon Ads MCP Server
The Amazon Ads MCP Server and Nova + MCP are complementary, not competing. Amazon's server is the execution layer for ads , create campaigns, adjust bids, change states. Nova's integration is the reasoning layer for the business , read the joined-up P&L, fees, PPC, inventory. Both can sit inside the same Claude conversation, with each tool registered to do what it is good at.
That separation is also a safety pattern. The campaign execution path goes through Amazon's own server, with Amazon's own permissions and rate limits. Nova MCP is scoped to the Nova workspace, reads by default and limits optional writes to listing edits that an owner enables and a user confirms. It never changes prices or stock levels.
Looking further out, the same MCP layer is what will let teams wire scheduled workflows on top of Nova (weekly P&L digests in Slack, fee-leak alerts, restock nudges) and build custom agents that combine the Nova data model with their own playbooks. Those remain roadmap items. Nova MCP currently reads by default, with owner-enabled listing edits previewed and confirmed before they reach Amazon.
What it means for sellers, agencies and aggregators
FBA sellers
Question-driven workflow
Instead of clicking through dashboards, ask the question. The answer cites the same data your dashboards already show.
Agencies
Faster client reviews
Weekly business reviews drafted from a prompt across the whole client book, grounded in the same Nova data the client sees.
Aggregators
Portfolio-wide questions
"Which 10 brands across the portfolio had the worst margin trend this month?" answered against the unified roll-up.
What to do now
- 1.
- 2.
Get your data model in order
If you are not on Nova yet, connect your account and load COGS. The AI is only as good as the data it sees.
- 3.
See your Amazon performance in Nova.
Try for free