Return store defaults an agent needs BEFORE creating products: regions & their currency codes, sales channels, stock locations, and price preferences. Call this first when unsure about currency_code / sales_channel_id / location_id.

List/search products. Supports free-text 'q', status filter, and filtering by collection/category/tag/type id.

Get a single product by id, including variants, options, prices, images and tags.

Create a product. Provide options (e.g. Size/Color) and variants that reference those options. Variant prices use major currency units (25 = $25.00). Images are passed as an array of public URLs. Set status 'published' to make it live, or 'draft' to stage it. Example variant: { title:'Medium', sku:'TL-M', options:{ Size:'M' }, prices:[{ amount:25, currency_code:'usd' }], manage_inventory:true }.

Update fields on an existing product (title, description, status, thumbnail, collection_id, etc.). Only include fields you want to change. To publish: set status='published'.

Delete a product by id. This is permanent — confirm with the user first.

Add a variant to an existing product. Provide options map + prices. e.g. options:{ Size:'XL' }, prices:[{amount:30,currency_code:'usd'}].

Update a variant (title, sku, prices, manage_inventory, options).

Delete a variant from a product.

List inventory items (optionally search by sku/title with 'q').

Set the stocked quantity for an inventory item at a stock location. Use list_inventory_items to find inventory_item_id and get_store_context for location_id.

List product collections.

Create a product collection.

List product categories.

Create a product category.

List product tags.

List product types.

List recent orders. Supports 'q' free-text and paging.

Get full details of one order by id.

List/search customers.

Low-level escape hatch: call ANY Medusa Admin API endpoint directly. Use for anything the curated tools don't cover (fulfillments, price lists, promotions, refunds, uploads, etc.). path must start with /admin. Be careful with POST/DELETE — confirm destructive actions with the user first.