Amazon Ads MCP Checker

An Amazon Ads MCP checker is a short readiness test, not a connection to your Amazon account. You name the AI client, the setup you have today, and the job you want. The result says whether that job works on the server today.

MCP means Model Context Protocol. Your assistant calls a server URL, and the server returns a tool result. Spillmark's URL is https://spillmark.com/mcp. Public tools answer calculator questions. Brand tools answer questions about one brand, and only with a key from Settings.

The checker doesn't ask for that key, an email, or a report upload. It also doesn't score Agent Central, DataDoe, or Amazon's own console. Those products have their own pages when the question is live writes.

The score is a count of jobs the server can do for the need you picked, out of 3.

1. Read-only profit, fee, and ACoS math scores 3. Connect the server and run the public calculators.
2. Geo-holdout and incrementality planning scores 2. Use the public holdout math, then a brand key if you want your own record.
3. Live Sponsored Ads bid and budget writes score 0, or 1 if you already have developer credentials. Credentials don't turn writes on.

A green result means connect https://spillmark.com/mcp. A red result means don't expect the assistant to change a bid through this server. The checklist under the score is the safety list: read-only versus write risk, a spend cap you keep in the ads console, and the fact that report uploads are not the same thing as an API credential.

Claude and ChatGPT have custom-connector steps on the Amazon Seller MCP page. The homepage lists Grok under Controlled by. This checker doesn't add a Grok setup that page doesn't document. Other MCP clients can try the same URL if they accept a remote server.

The example is Claude, a seller account with CSV reports only, and a need for read-only profit, fee, and ACoS math. The status is ready. The endpoint is https://spillmark.com/mcp. No Ads API key is required.

Change the need to live bid and budget writes and the status flips, even if you also pick developer credentials. The credentials note is there so a developer doesn't confuse "I can call Amazon's API" with "this server will call it for me." It won't.

Change the need to holdout planning and the result points at the holdout test calculator. That calculator uses order counts you type. It doesn't read a brand until you connect one with a key.

A good setup is the one that matches a live tool:

• You want fee, profit, ACoS, or TACoS math, and you can type the inputs or upload reports to the dashboard.
• You want to know if a geo holdout can see a realistic lift before anyone pauses ads.
• You already use Claude or ChatGPT and can paste a server URL as a custom connector.
• You will keep the brand key out of a shared chat. Owners create keys in Settings and can revoke them.

From the result, open Amazon Ads MCP for the server explanation, the ACoS calculator for a single product's break-even check, and how lift is calculated if the next question is cause rather than credit.

Spillmark plans geo holdout tests for Meta ads across US states from your Amazon order history. The dashboard is free. Standard is $49 a month for one Amazon account, with holdouts you approve. You get a 7-day trial for $0.

The usual miss is treating a roadmap line as a button.

• Asking the assistant to raise a budget tonight. The server can't send that change.
• Assuming a Seller Central password or a CSV export is an Ads API credential. The public calculators don't need either.
• Pasting a brand key into a client you share with a contractor. The key reads one brand. Revoke it if the chat is no longer yours.
• Skipping a spend cap in the ads console because a future autopilot might have one. The cap is yours until write tools exist.

If live writes are the job, read the Agent Central alternative and the DataDoe alternative. Both ship Amazon write-back today. This server doesn't, and the checker says so instead of offering a fake connect button.

A lot of Amazon Ads MCP searches are about bids. The measurement job is different. Platform ACoS is credit. A geo holdout is one way to see what you would lose if the ads stopped. Spillmark runs geo holdout tests for Meta ads and can plan a branded Amazon PPC test you approve. The design locks before the test starts.

The checker's middle need is that planning job. It links the holdout calculator so you can see the smallest lift a test could detect before you pause a state. Geo testing walks through the design if you want the longer explanation.

Read-only access is the point. An assistant can see the range and can't change the budget while the test runs.