Sellufy now runs an MCP server. Connect it to Claude Code, claude.ai, Cursor or another AI app, and that app can find and change the products in your Sellufy catalog while you talk to it.
Type “enrich my 5 newest shoe products, casual tone” into Claude Code. Claude looks up the products, runs Sellufy’s AI enrichment on them, and the new titles, descriptions, categories, tags and attributes land on the product records. You don’t copy anything into the chat or paste anything back out.
It comes with every paid plan, starting with Growth at $9 a month.
What a PIM MCP server does
MCP stands for Model Context Protocol. Anthropic published it in November 2024 as an open standard for connecting AI apps to outside tools. Claude, Cursor, VS Code, Gemini CLI and Codex all support it.
An MCP server gives the AI app a list of tools and what each one takes. The app reads that list when a chat starts. When you ask for something, the app picks the tools to call, Sellufy runs them against your catalog, and the app reads the result.
A PIM is where your product records live before they go out to Shopify, WooCommerce or anywhere else. Putting an MCP server on it means the AI works on those records directly. Before, AI on a catalog meant exporting a CSV, pasting rows into a chat, and copying the answers back one field at a time. That works for a handful of products. Past a few dozen, descriptions start landing on the wrong rows.
The 12 tools
Five of them only read.
| Tool | What it does |
|---|---|
list_products | Searches products and filters by status, brand, category or tag |
get_product | Returns one product in full |
list_taxonomy | Lists brands, categories, tags, attributes or options, with their ids |
list_assets | Lists files in the asset library |
get_job | Returns the status of a background job |
create_product | Creates a simple product |
update_product | Changes fields on one product |
bulk_update_products | Sets the same status, brand, categories, tags, price or stock on up to 100 products |
delete_product | Deletes one product |
enrich_products | Rewrites title and description and fills in taxonomy and attributes, 10 at a time |
generate_product | Creates up to 10 products from a text prompt |
generate_product_images | Draws a product image and adds it to the product |
enrich_products and generate_product do the same work as the AI enrich and AI generate buttons
in the dashboard. You ask in a chat instead of clicking.
Things to ask it
The AI app chains the tools on its own. You describe the result you want.
- “Add the Summer tag to every product with linen in the title.”
- “Move every drafted product in the Mugs category to published.”
- “Enrich the 10 newest products in Footwear. Short, casual tone.”
- “Create 5 products for a ceramic pour-over coffee set, then draw an image for each one.”
- “Set stock to 0 on everything tagged Discontinued.”
For the linen request, Claude calls list_products to find the products, list_taxonomy to get the
id of the Summer tag, then bulk_update_products once for all of them.
Who pays for the AI
An MCP session runs two kinds of AI, and you pay for them in different places.
The chat runs on your AI app. Whatever you pay for Claude, Cursor or Codex covers it. Sellufy never sees that bill.
Three tools also call an AI model from inside Sellufy. Those are enrich_products,
generate_product and generate_product_images, and they run on an API key from your own AI
provider that you save in Sellufy. They never spend Sellufy AI credits. The AI buttons in the
dashboard still use credits, the same as before.
For text you can pick OpenAI, Anthropic, Google Gemini, DeepSeek, or any OpenAI-compatible API such as OpenRouter, Groq, Mistral or xAI. The model has to support tool calling. For images you can pick OpenAI, Google, Stability AI, Replicate or fal.
Only the account owner can save the key. Sellufy makes one small test call with it, saves it only if that call works, and stores it encrypted. After that you only ever see the last 4 characters.
The AI cost sits on your key, so we don’t meter MCP by plan. Growth gets the same 12 tools as Professional.
What a key can do
Each MCP key belongs to one person and one workspace, and it acts as that person. If your role can’t delete products in the dashboard, your key can’t delete them either.
A read-only key hides the 7 write tools. The AI app only ever sees list_products, get_product,
list_taxonomy, list_assets and get_job. You can switch a key between read only and read and
write in Settings, and the key itself doesn’t change, so the AI app needs no new setup. Start a new
chat after you switch it. AI apps load the tool list when a chat begins.
Sellufy shows a key once, when you create it, and stores only a hash of it. If you lose a key, revoke it and make a new one. Each person can have up to 20.
The usual limits still apply. MCP requests count toward your plan’s rate limit. A tool that would take you past a plan quota, such as your product count, returns the reason to the AI app. One call enriches or generates at most 10 products, so a request for 40 becomes 4 calls.
How to connect it
- In Sellufy, open Settings, then MCP, and click Create key. Name it after the app it’s for, pick the workspace, and turn on Read only if you want that.
- Copy the key. The same dialog has a ready-made config for each AI app.
- Add the server to your AI app.
Claude Code takes one command:
claude mcp add --transport http sellufy https://api.sellufy.com/v1/mcp --header "Authorization: Bearer sk_mcp_..."
In claude.ai, open Settings, Connectors, Add custom connector and enter
https://api.sellufy.com/v1/mcp. Under Authentication pick No sign-in, then add a header
named x-api-key with your key as the value. claude.ai only sends a short list of approved header
names, and Authorization isn’t on it.
Cursor, VS Code, Gemini CLI and Codex each take a config file with the same URL and an
Authorization header. The MCP docs have the exact file
for each one.
When to use it and when not to
If you already spend part of your day in Claude or Cursor, the MCP server saves trips to the dashboard for the small jobs, like retagging 30 products or filling in categories after an import.
For one product where you want to see every field and every image side by side, open the dashboard. A chat is a poor place to review a product record.