Tools

The server exposes 42 tools: 38 generated from the public API — one per operation, named after its operation id — and 4 hand-written workflow tools that chain calls together. You don’t need to memorise these; the model picks them. The list is here so you can tell what it did, and approve or block individual tools in clients that support it.

​

Workflow tools

Shortcuts for the questions people actually ask. Each takes a single brand_id and returns the already-joined result, so the model spends one round-trip instead of three.

​

Generated tools

Every tool takes the same parameters as the underlying endpoint: path ids as plain arguments, and the query parameters (page, page_size, sort, filter[…]) as optional ones.

​

Brands

​

Brand metrics

​

Topics

​

Prompts

​

Citations

​

Competitors

​

Ads & advertisers

​

Reading the results

Tools return the API’s data payload. That means the metric units apply verbatim — including the trap that sentiment is a 1–100 average rather than a percentage. If an assistant reports sentiment as “62%”, it has mislabelled the unit; the number itself is right.

​

Writes

Eleven of these tools mutate data: create_brand, update_brand, delete_brand, create_brand_topics, update_brand_topic, delete_brand_topic, create_brand_prompts, update_brand_prompts, and the three competitor writes. Deletes are soft — the row drops out of the lists but its history is kept, and an archived prompt can still be read directly.