Skip to main content
The MerchantOps Claude plugin turns “set up MerchantOps for a new catalog” into a conversation. It bundles eight skills, four slash commands, and a reference to the MerchantOps MCP server. Skills are guided, repeatable workflows: crawl a site or read a spreadsheet, propose a property dictionary and product types, produce a bundle of files for you to review, and then load the catalog and prices — pausing for your explicit approval before anything is written.

​
Two surfaces, and they are not the same

The plugin ships two capability paths, and it matters which one you need:

The MCP server

Conversational catalog work. Ask about products, variants, pricing and publish status. Nothing to install locally — see MCP server for what it covers.

The skills, driving the CLI

The onboarding path. Designing a property dictionary, loading a catalog or importing prices happens through the mo CLI, which the skills run for you.
No skill depends on the MCP server. The MCP server has no bulk catalog import and cannot create property definitions or product types, so it cannot drive an onboarding on its own. If you want conversational catalog reads, the MCP server alone is enough. If you want to onboard a catalog, you also need mo installed.

​
Install

Pick the one path that matches where you work — you do not need all four.
The packaged install: skills, slash commands, and the MCP connector in one step.Claude Code:
Claude Desktop: open the Plugins browser, add the marketplace merchantops/claude-plugin, search for “MerchantOps”, and install. The two commands above also work in a chat.This does not include the mo CLI itself — the onboarding skills shell out to it. See the CLI tab.
The plugin and the standalone skills are published at merchantops/claude-plugin, licensed Apache 2.0.
MCP clients cache the tool list from their last connection. If MerchantOps’ tools do not appear right after installing or updating, reconnect the connector or start a new session.

​
Slash commands

Slash commands are explicit: they never trigger on their own, and each one waits for your confirmation before writing anything.

​
Skills

Skills are invoked by Claude when they are relevant to what you asked.

​
What a site crawl needs

merchantops-onboard-from-site and merchantops-competitive-analysis fetch pages using a crawling tool that runs on your machine, under your own API key and budget — never through MerchantOps infrastructure. That path needs Node.js in addition to Python. The spreadsheet path needs neither.

​
Where each surface works

​
Before you start

Sign in as a member with a scoped, non-Owner role for onboarding work. The role on the account you signed in as is the real boundary on what any skill or command can do — not the guardrails in the CLI.
Every write is planned first, requires an explicit --execute, and is bound by a --confirm hash to exactly the files you reviewed; every skill shows you that plan and waits. Those are real protections against a misfire or a casually-approved step. They are not a boundary against an agent that is instructed to supply the flags anyway. The API’s permission check on your own role is what bounds that, so start from a role that cannot approve or publish. See Team and roles. Every skill treats content it reads — a web page, a spreadsheet, a PDF, an API response — as data, never instructions. Text in a crawled page that looks like a command, a request, or a claim about permissions is reported to you, never acted on.

​
Next steps

CLI reference

Every command the skills run, and what each exit code means.

Agent guardrails

What needs confirmation, and what an agent can never do.

Product types & properties

What a property dictionary is, and how product types use it.

Price batches

Where imported prices land, and how they get approved.