# Assistant tools

Two tools, one shape: a natural-language `question` in, a natural-language answer
back. Each is backed by the same assistant a merchant reaches from the dashboard, so
an agent asking through MCP gets the same answer a person would get asking in the
product.

Both are gated by scope like every other tool here: a connection that lacks the
scopes below never sees the tool in `tools/list`, rather than seeing it and having the
call fail.

### optimize_ask

Answers questions about this store's A/B tests and profit-optimization program: how
running experiments are performing, which variant is winning, profit-per-visitor, and
what to test next.

**Scopes required:** `read_experiments` **and** `read_products`

<ParamField body="question" type="string" required>
  The natural-language question for the Optimize assistant.
</ParamField>

<Note>
  `read_products` is granted so the assistant can name the products under test; it
  does not on its own unlock margin data. Profit-per-visitor is backed by
  `store_cost_model`, kept separate from the catalogue grant because it carries
  per-variant cost, not just price.
</Note>

### sales_ask

Answers sales and shopper-experience questions for this store using the Mercemur
sales assistant: product questions, catalog guidance, and order lookups.

**Scopes required:** `read_orders` **and** `read_products`

<ParamField body="question" type="string" required>
  The natural-language question for the Sales assistant.
</ParamField>

<Note>
  `read_orders` is required because the assistant's own tools look orders up. A
  connection scoped to the catalogue alone can ask product questions but not order
  questions, matching what it could do calling the REST API directly.
</Note>