Automate Vendor Listing Quality at Catalog Scale

Yes. A Vendor Central MCP Skill can rank catalog gaps by page views, resolve each product type schema with Amazon Agent Atlas, generate compliant imagery through Amazon Agent Iris when needed, and hold validated writes for approval. Amazon already gives you three ways to change a listing. The hard part is deciding which attributes matter, what a correct value looks like, and which SKUs deserve attention first.

The ranked audit appears before a write is proposed:

That distinction matters because a bulk spreadsheet solves only the upload. Someone still has to identify the right product type, resolve its recommended attributes, and provide a valid value for every row. The mechanism is not the bottleneck.

Why not ask ChatGPT or Claude? A plain chat hits the same three walls on any Amazon job. ChatGPT and Claude have no access to your data, so they cannot see which SKUs carry a listing quality opportunity or what the detail pages say today. They have no way to take action, so they cannot write an attribute or upload a feed; the most they can do is hand you values to apply yourself. And they run on generic knowledge, not Amazon's, so they do not know which attributes Amazon currently recommends for each product type. You get disconnected, generic, manual work, applied one panel at a time. The Vendor Central MCP provides your catalog and the tools to act, Skills provide the repeatable workflow, and Atlas supplies the Amazon rule book.

The Listing Quality Indicator grades two pillars, and both vary by product type

Amazon's own definition is precise. The Listing Quality Indicator measures how complete and correct a detail page is across content quality, which evaluates product data such as titles, bullet points and key attributes for correctness, completeness and policy compliance, and multimedia quality, which reviews image and A+ content quality and completeness.

Two things follow that most catalog projects miss. First, requirements vary by product type, so there is no single checklist to apply across a catalog. Amazon publishes a per-product-type recommended attribute list and updates it monthly. Second, the score is explicitly a directional guide: it compares your listing to the strongest listings in similar categories and does not account for dynamic factors like offer price, fulfilment or seasonality.

The Improve listing quality and Enhance listings experiences surface these opportunities per ASIN, with page views available as a prioritisation signal. They are related but not identical: Improve listing quality changes can take up to one and a half hours to appear, while Enhance listings refreshes its quality indicator within 15 minutes after you publish updates.

Amazon gives you three ways to write the change, and none of them decide what to write

This is the claim worth internalising before buying anything. The three documented paths are all mechanisms:

The API path is the one people underestimate. Amazon states it plainly: register a developer profile, create the required AWS resources and Selling Partner API (SP-API) application, develop the functionality, test the application, then release and maintain.

So the honest options are click forever or fund an engineering project. An agent on the Vendor Central MCP is a third: the same APIs, your account, no application of your own to maintain. Kuudo also exposes shared SP-API operations through the Selling Partner MCP for workflows that span vendor and seller accounts.

Product Type Definitions turns "recommended" into a schema the agent can satisfy

The foundational API is the one nobody markets. Product Type Definitions lets you search for available product types by marketplace and keyword, then retrieve the definitions and schemas describing the attributes and data requirements for that type.

The important property is reuse: those same schemas back the Listings Items API, the Catalog Items API and the bulk JSON_LISTINGS_FEED. Validating against the schema catches attribute and data-requirement errors before submission. It does not guarantee publication, because Amazon can still surface processing, contribution, or policy issues afterward. That is the difference between an audit that merely proposes edits and one that reduces preventable submission failures.

The read side pairs with it. The Catalog Items API returns reconciled information for an ASIN, equivalent to what the detail page displays, including images, product type and item identifiers. Read the current state, resolve the schema, then compute the gap.

The cold-open table is the artifact. It ranks the gaps using the signal Amazon itself offers for prioritisation, and every content recommendation is tied to the product type schema that produced it.

Image Manager moves files; it does not know which of Amazon's rules apply to this SKU

Multimedia is one of the Listing Quality Indicator's two pillars, so a catalog project that only fills text attributes leaves image and A+ gaps untouched. But Manage Images is a tool for a manual process. Image Manager lets you upload individual images, delete them, copy a media set to sibling products, and preview whether they are live. It moves and organises files. Everything that decides whether the file was the right one happens somewhere else.

Amazon is direct about the consequence: uploading an image does not guarantee it will be displayed, because Amazon selects and arranges images from multiple sellers, and the way to improve your odds is to comply with the product image requirements. Compliance is evaluated after you upload, not before. So the manual loop is produce the file elsewhere, upload it, wait up to 24 hours for display, and if it was non-compliant, learn that from Image suppression and start over. The troubleshooting path even depends on which submission method you used.

The harder problem is that there is no single rulebook to learn. Image requirements in the amazon_vendors corpus are category-specific: Multipack imaging standards requires the main image to show the total quantity delivered and the product in its packaging, with the title carrying that count as well. Technical image file requirements accept RGB and mark CMYK as not preferred, with instructions to convert it to RGB. Replacing a suppressed main image means uploading with the MAIN variant code rather than deleting the old one, because every ASIN requires a main image.

The sharpest example is the newest one. How to tag media that contains an AI-generated person exists because some jurisdictions require disclosure, and Amazon requires the media to be tagged before upload. Its documented procedure is to open the file in Preview on macOS, press Cmd+I, and add a keyword in the Keywords tab, or on Windows to use the Tags field in File Explorer, with an explicit warning that Subject and Tags write to different metadata locations. That is the state of the art for compliance on AI-generated imagery: a per-file manual metadata edit, done correctly, every time, across a catalog.

An agent inverts the order of all of this. It reads the current image state through the Catalog Items API, resolves which of those rule sets apply to this product type from Atlas before anything is produced, generates through Amazon Agent Iris against live product data rather than a stock template, applies the disclosure metadata where the rule is triggered, and writes through the Listings Items API with the correct variant code, validated against the product type schema. The multipack case shows why that ordering matters: the fix is an image and a title together, and doing one without the other leaves the listing non-compliant in a new way.

The part that is yours rather than Amazon's matters just as much. Amazon's requirements are public and identical for every vendor. Your brand rules are not. Atlas grounds the Amazon rules; your private style guide supplies the palette, framing, voice, and tone. The Skill enforces both in the same pass, so compliant and on-brand stop being separate review cycles.

What happens next

The audit is worth running once. The loop is worth keeping. The Notifications API sends LISTINGS_ITEM_ISSUES_CHANGE whenever the issues attached to a SKU you own change, and those issues are exactly the ones that can push a listing to inactive or search suppressed. Subscribed through the Amazon Agent Data layer, that notification is the trigger: a SKU develops a problem, the agent re-resolves the schema, proposes the fix, and waits for approval before writing.

Saved as a reusable Skill, the same loop runs against the whole catalog on a schedule and composes with the vendor reporting work you already do, such as the ARA report and metric glossary. Read and written through the Vendor Central MCP, it sits alongside the Amazon Ads MCP when the question turns to what the traffic did once the detail page was complete. One Amazon Agent Data layer, grounded by Atlas throughout.

The mechanism was never the hard part. Deciding what to write, per product type, across a catalog nobody has time to read, is the hard part, and it is the part that finally has an operator.

Next in this series: the same catalog-scale review loop applied to A+ content.