The Amazon Ads MCP Is Real. Here’s What It Can’t See.

On February 2, Amazon released an open beta for the Amazon Ads MCP Server. If you missed it: Model Context Protocol is a standard that lets AI tools — Claude, ChatGPT, Gemini — talk directly to external services. In this case, your Amazon Ads account.

Coverage has been breathless. “AI can now run your campaigns.” “Hands-free advertising.” I get why it’s exciting. But I’ve been watching this closely, and I think the more useful conversation is about what the AI can’t see — not what it can do.

The MCP is a connection layer. Instead of logging into Seller Central to manage campaigns, you describe what you want in plain English, and an AI handles the API calls. Create a campaign, adjust bids, pull a performance report — all through a prompt interface.

It’s not Amazon’s own Ads Agent, which lives inside the Amazon console. The MCP is different: it’s an open protocol that lets external AI tools connect to your ad account. When you use Claude or ChatGPT with the MCP, you’re giving that AI direct API access to your ad data and spend.

That distinction matters. Amazon controls what their in-house agent can see and do. The MCP opens the same door to any AI tool you choose to connect — which is both the power and the risk.

Alex Willen at The Automated Operator ran one of the more rigorous real-world tests I’ve seen. He connected Claude to his ad account and let it run. His revenue doubled. That’s a real result and worth taking seriously.

There’s a catch in the methodology worth noting: Claude was trained on his Black Friday and Christmas data — the highest-velocity, highest-conversion period of the year. Applying those patterns to a May campaign means you’re essentially using seasonal performance as a baseline for normal weeks. The numbers will look different.

That’s not a knock on the tool. It’s a reminder that AI optimizes what it can see. If the training window is skewed, the optimization will be too.

Here’s the part that didn’t get enough coverage: during internal testing, Amazon’s own AI went off-script. It made decisions outside what Amazon intended. Their response was to build guardrails — forced API paths that block anything outside approved scope.

They also introduced the BSA Agent Policy, effective March 4, 2026. It requires documented human authorization for any automated price change exceeding 20% within a 24-hour window. The policy exists because automated agents can move faster than any human can catch — and when they move in the wrong direction, the damage compounds quickly.

I’d read about repricers going rogue with sellers for years. Watching Amazon face the same problem with their own internal tooling was instructive. It’s a genuinely hard problem. Speed and human control are in tension by design.

The MCP has access to your advertising data: campaigns, bids, keywords, ACoS, spend. That’s useful. But it’s a narrow slice of what actually determines whether a campaign should be running, or running harder.

Inventory. If you’re three weeks from a stockout, scaling a campaign isn’t the right call. The AI doesn’t know that.

Your actual margins. It knows your advertising cost of sale. It doesn’t know your landed cost, your FBA fees, or whether the margin left after ad spend is worth the volume.

Buy Box status. If a competitor holds your Buy Box on a listing, you’re paying for clicks that will convert for them, not you. The MCP has no visibility into Buy Box share on your ASINs.

Organic ranking data. Advertising and organic search interact, but the MCP can’t see your organic position or how it’s trending.

This isn’t a criticism of the tool. It’s the nature of what the Ads API exposes. But any optimization built only on that data is optimizing a partial picture.

The MCP can make your campaigns more efficient within the advertising data it can see. What it can’t do is tell you whether you’re bidding on the right listings in the first place.

Most sellers who ask me about Amazon automation are thinking about speed — how fast can the tool act, how often does it update. That’s the wrong question. The right question is: what does the tool need to know before it acts?

Spend goes to waste in two main ways on Amazon. One is inefficient bidding — the AI is already good at fixing that. The other is bidding on listings with structural problems: Buy Box competition you’re losing, suppressed listings, inventory gaps that make the traffic meaningless. Optimizing the first without addressing the second is like tuning an engine when the wheels aren’t on the ground.

Connecting advertising performance to what’s happening with your Buy Box, your inventory, and your margins — that’s still yours to manage. No AI tool currently bridges those signals automatically.

I built SentryKit because that gap existed before MCP was a concept. When I was selling on Amazon, I kept running into situations where my ad performance looked fine but something was wrong upstream — a competitor had taken my Buy Box, a listing had been suppressed, a stockout was coming. The advertising data didn’t surface any of it. I needed a different layer of intelligence.

SentryKit is a Buy Box intelligence platform, not an advertising tool. But for sellers using MCP-based ad automation, the two are complementary: knowing your competitive position before the AI runs your campaigns gives the automation something solid to build on.

If I were evaluating the Amazon Ads MCP today, here’s where I’d focus.

Start with your healthiest listings — ones where you hold the Buy Box consistently, inventory is stable, and margins are clear. That’s the environment where MCP optimization will have the cleanest signal. Listings with Buy Box competition or inventory variability introduce noise that the AI isn’t equipped to filter.

Build a human review layer. The BSA Agent Policy requires it for large price moves, but even for bid adjustments, knowing what the AI did and why is worth tracking. “Claude handled it” is not a useful post-mortem.

Don’t train on seasonal data if you’re running normal-week campaigns. If the training window is Black Friday, the output will be calibrated for Black Friday.

And monitor your Buy Box share on anything you’re running ads against. If a competitor is holding your Buy Box and you’re not watching it, you’re funding their conversions.