Catalog Attribute Normalizer — Claude Code MCP server: what it does
Turns messy product attributes from several sources into one canonical form and maps categories against real Google and Shopify taxonomies.
claude mcp add --transport http catalog-normalizer https://catalog-normalizer.acjlabs.com/mcpWhat Catalog Attribute Normalizer does
The server takes a batch of products with title, description and raw_attributes, then hands back cleaned-up values: sizes, colors, units. Category paths come from retrieval, not recall. Candidates are pulled out of the live Google Product Taxonomy and Shopify files and shown to the model as options, so a returned leaf_id usually points at a node that exists. Amazon gets no such backing, since there is no public reference file to retrieve from, and stays best-effort.
Attribute values are safe by construction: they come from what you sent and only get canonicalized. When the model infers a value from the title or description, provenance reads "extracted" — treat that as a hint. The vocabulary is fixed (size, color, material, gender, sleeve_length); anything else is dropped rather than renamed.
It runs as an MCP server exposing the normalize_catalog tool, or as a plain HTTPS API. You issue your own key with POST /v1/signup and an email in the body; the key comes back in the response and is shown once, so losing it means signing up again. The free tier covers 500 products a month. Billing counts products, not calls — a batch of 40 spends 40. A batch is all-or-nothing: if it exceeds what is left on your plan, the request is rejected with 402 and nothing gets classified.
What it helps with
- reconcile size and color spellings across a Shopify export, a supplier CSV and a marketplace scrape
- get a category with a verified
leaf_idinstead of one the model made up - canonicalize units before pushing products into a feed
- tell a value taken from your input apart from one the model inferred
- tag a batch of products for both Google and Shopify taxonomies at once
Example prompts for the agent
- Normalize the attributes in this product batch and map categories for google and shopify
- Run these
raw_attributesthroughnormalize_catalogand flag every value withprovenanceset toextracted - List the results where
leaf_idcame back null and explain why
How to install Catalog Attribute Normalizer
Platform CLI: remote server
claude mcp add --transport http catalog-normalizer https://catalog-normalizer.acjlabs.com/mcpPlatform CLI: remote server
codex mcp add catalog-attribute-normalizer --url https://catalog-normalizer.acjlabs.com/mcpConfig file
{
"mcpServers": {
"catalog-attribute-normalizer": {
"url": "https://catalog-normalizer.acjlabs.com/mcp"
}
}
}Config file
{
"servers": {
"catalog-attribute-normalizer": {
"type": "http",
"url": "https://catalog-normalizer.acjlabs.com/mcp"
}
}
}Platform CLI: remote server
gemini mcp add -t http catalog-attribute-normalizer https://catalog-normalizer.acjlabs.com/mcpStep by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about Catalog Attribute Normalizer
How do I install Catalog Attribute Normalizer in Claude Code?
Run claude mcp add --transport http catalog-normalizer https://catalog-normalizer.acjlabs.com/mcp. Add an Authorization: Bearer YOUR_API_KEY header to authenticate.
What does Catalog Attribute Normalizer do?
It takes a batch of products and returns canonical attributes plus category paths for the taxonomies you ask for. It works as an MCP server or as an HTTPS API.
How is attribute normalization different from category classification?
Attributes are canonicalized from data you already supplied, so there is no recall involved. Categories rely on retrieving candidates from taxonomy files, and Amazon has no such file — results there are weaker.
