At 1:27 PM on Wednesday, September 23, 2026, a Slack message comes in from a teammate sitting in the audience at Amazon Accelerate. It is a photo of the keynote screen, huge red letters across the stage: “Selling partner plugin.” Under the photo, two words.
“Not great.”
Two minutes later, the announcement follows. An hour after that, the Claude directory listing, and a screenshot of a waitlist confirmation: “You’re on the list! We’ll let you know when it’s your turn.”
Another teammate replies: “That is definitely not great.”
I read the announcement twice. Amazon has taken Seller Assistant, the AI inside Seller Central, and plugged it into Anthropic’s Claude. Listings, inventory, sales analytics and real-time metrics, all in a chat window. It recommends price changes and listing edits and, with the seller’s approval, makes them. It remembers the seller’s pricing patterns and inventory cycles from one session to the next. It runs workflows in the background while the seller sleeps. Amazon says connecting Claude takes about 60 seconds, with no coding.
And it’s free.
Seller Labs sells an MCP server that connects a seller’s Amazon data to Claude or ChatGPT. So for the rest of that afternoon, the question in my head was the one on the thumbnail of this post. Did Amazon just kill us?
What Amazon actually launched
Amazon announced the Selling Partner plugin at Accelerate on September 23, 2026. At the time of writing:
- It is live in Amazon Quick and in beta in Claude, for sellers in Amazon’s U.S. stores. Access to the Claude beta opens in waves from a waitlist.
- Actions need the seller’s approval and leave an audit trail. Sellers choose whether a workflow only recommends or also executes.
- Seller Assistant keeps a persistent memory of each seller’s pricing patterns, inventory cycles and growth goals, and that memory follows the seller across Seller Central, Quick and Claude.
- Always-on workflows keep watching while the seller is offline: a top product dropping below four stars, competitor pricing, inventory thresholds.
- Every primary account holder gets 12 months of Quick Plus free, through December 31, 2026.
Amazon also shared two numbers. Sellers accept Seller Assistant’s recommendations more than 90% of the time, and 90% of Amazon’s selling partners already use third-party AI tools.
That second number is why this exists. Nine out of ten sellers were already carrying their Amazon data somewhere else to think about it. Amazon decided to meet them there. Mary Beth Westmoreland, Amazon’s VP of Worldwide Selling Partner Experience, put it plainly to GeekWire: “Our vision was that they would never have to log into Seller Central. We would just bring it to them where they work.”
It is a good product. I expect it to get better every quarter.
Free isn’t free
By that evening I had stopped asking whether it was good, and started asking who it works for.
On every Amazon seller’s business, Amazon is the landlord, the fee collector, the ad seller and the referee. It sets the referral fees, the FBA fees and the storage fees. It sells the Sponsored Products placements. And with this launch, it is also the AI advisor that helps the seller decide what to do about all of it.
I’m not saying the advice will be bad. I’m saying the advisor sits on the other side of the table. When the question is “should I spend more on ads”, “are these fees right” or “should I send in more inventory”, the party answering gets paid either way.
Would you let your landlord do your books?
The 90% acceptance rate cuts both ways. It can mean the recommendations are good. It can also mean that by the fiftieth approval prompt, most people stop reading them.
“Free” still has a price. With Amazon’s assistant, the seller pays it in two currencies.
Context. Real profit advice needs your landed cost: COGS, freight, duties, the numbers that decide whether a SKU makes money at all. Amazon only has them if you hand them over. Seller Assistant’s persistent memory of your pricing patterns and growth goals is convenient, and it is also your business logic, held by the counterparty that sets your fees.
Limits. Amazon decides what its plugin can see, which actions it can take and what its recommendations favor. At the time of writing, the plugin’s announced scope covers listings, inventory, metrics and sales analytics. Running ad campaigns, bids and keywords is not on the list.
And the free Quick Plus year ends on a date. Amazon hasn’t said what it costs after that.
What changed my mind
On September 23 I thought our edge was connecting your Amazon data to Claude and ChatGPT. By that night I knew Amazon had just made the Claude half free, others had already made it common, and our job is to be the one AI advisor on your side of the table.
I won’t pretend the basic layer, “chat with your Seller Central”, is still a moat. It isn’t, and not only because of Amazon: plenty of other companies have shipped an Amazon MCP too. Connecting Amazon data to Claude or ChatGPT is table stakes. What matters is what the AI finds when it gets there, and that comes down to three things Amazon’s plugin isn’t built to do.
1. Data built for an AI to reason over. Hand an AI raw Amazon reports and it has to guess: which report holds which fee, how refunds net against sales, whether a conversion rate gets averaged or summed. It guesses wrong often enough to matter. The Seller Labs MCP Server runs on Data Hub, which does that work before the AI arrives. Each seller’s data lands in their own dedicated database of more than 50 tables: orders, settlements, fees, ads, inventory, returns, reimbursements, search queries and rank. The hardest join is already done. Every SKU gets one profit row per day, with sales, refunds, FBA and referral fees, storage, aged-inventory surcharges, Sponsored Products spend and your COGS, across more than two years of history, for every marketplace on your account. Agencies go one step further and add each client’s seller account to the same connection, so one question can run across every brand they manage. Next to the data sits a map written for the AI: which table answers which question, how to calculate each metric correctly, and a cookbook of tested queries. Ask “which keywords are spending money on products that lose money?” and Claude follows a path from keyword to ad to SKU profit that has already been worked out, instead of improvising one.
2. Ads you can actually run. Through the same connection, Claude can build a Sponsored Products campaign, add keywords and negatives, change bids, and pause or re-budget Sponsored Brands and Display campaigns. Anything that spends real money stops and asks first, and the guardrails know your business: a warning before launching ads on a product with less than 14 days of stock, or before setting a budget more than double your recent daily spend.
3. Your context stays yours. Your COGS and margins live in your own dedicated database, not with the party that sets your fees. You decide what the agent sees and what it is allowed to do. Seller Labs doesn’t collect your referral fees or sell you ad placements.
Move upstream: agents that run the business
Amazon’s always-on workflows point in the right direction: AI that keeps working after you log off. The question is what that agent can see when it wakes up.
An agent that sees only Seller Central can tell you a product dropped below four stars. An agent that sees the whole P&L can do the Monday morning work for you:
- Pull last week’s profit by SKU and flag anything that went negative.
- Find keywords that spent without converting, and queue negatives for your approval.
- Check days of cover against each product’s lead time.
- Project the cash the next inbound order will need.
That is the move upstream. The job used to be connecting your data. The job is becoming running your business with agents that work for you: ads, inventory and cashflow planning, on a schedule, with your approval on anything that spends money.
You don’t need to be technical to set this up. Point Claude Code or Codex at the Seller Labs public toolbox on GitHub, which includes a task scheduler (Windows only, at the time of writing), and tell it to set you up. The instructions are written for your coding agent to follow.
Amazon’s plugin vs. an independent MCP
At the time of writing:
| Amazon Selling Partner plugin | Seller Labs MCP Server | |
|---|---|---|
| Price | Free in beta; Quick Plus free through December 31, 2026, price after that unannounced | Paid subscription |
| Who it works for | The marketplace that sets your fees and sells your ads | You |
| Where your costs and business logic live | With Amazon, if you hand them over | Your own dedicated database |
| True profit (COGS, landed cost) | Only what Amazon holds | Daily profit by SKU with your COGS |
| Data shaped for AI reasoning | Not described in the announcement | 50+ tables, one profit row per SKU per day, query map and tested cookbook |
| Ad campaigns, bids, keywords | Not in the announced scope | Create, bid, negate, pause, re-budget |
| Always-on / scheduled work | Amazon’s workflows, Amazon’s limits | Your agents, your schedule, your rules |
| Works with | Amazon Quick; Claude in beta | Claude and ChatGPT |
| Availability | U.S. stores, Claude beta by waitlist | Every marketplace on your account |
| Multiple seller accounts | One account and marketplace at a time, per early reports | Agencies add each client account to one connection |
So, did Amazon kill us?
No. It took the easy part of what we sold, and it made the rest of it more obvious.
If you run a seven- or eight-figure Amazon business and you already work in Claude or ChatGPT, try this: connect the Seller Labs MCP Server to it. If you work in Claude, set up one scheduled agent too. Then ask it the question Amazon’s assistant isn’t built to answer: which of my SKUs lost money last week after ads, fees and COGS?
Use Amazon’s assistant. It’s good. Just make sure the one you trust with your margins is sitting on your side of the table.
