Skip to main content
A Vertical is a curated, domain-specific intelligence layer exposed over MCP. Where your own data tables hold your orders, customers, and leads, a Vertical holds the market around you — the entities that matter in your industry and everything known about them. It is not a generic web search and not a raw database query. It is a deliberately modeled surface:
  • Canonical entities — the brands, models, or players of the domain, each with a stable entity_id.
  • Structured signals — named, typed observations attached to an entity (growth, paid-media, engagement, launches, reviews, …).
  • Time-based observations — each signal is anchored to a period, so you can see change, not just a snapshot.
  • Cross-entity comparison — aligned signals across several entities in one call.
  • Market benchmarks — the “what is normal” aggregates you read a single entity against.
  • Reference lookups — the controlled vocabulary (countries, tiers, segments, platforms) the surface uses.
  • Editorial know-how — guides, playbooks, and interpretation rules that explain what the numbers mean.
  • Evidence links — public source references behind a signal, on request.

​
The core principle: start from the business question

The most important habit when using a Vertical: start from the business question, not the signal taxonomy. The AI maps a natural-language intent to the right tool and the right signals. Signal names (growth_30_60_90, omni_channel_score, trend_break, …) are an agent-facing vocabulary — an internal index the AI uses to fetch the right evidence. They are not something the user has to know or name.
Example. A user asks “Is this brand growing?” — the AI internally reaches for the growth signals, retrieves them, and answers in plain language. The user never needs to know a signal called growth_trajectory_class exists. The intent drives the tool selection; the taxonomy stays behind the curtain.
This is what separates a Vertical from a search box. A search box makes the user speak its language. A Vertical lets the user speak in business terms, and the AI does the translation into signals and tools.

​
What a Vertical is not

  • Not a generic web search — it returns curated, benchmarked signals, not arbitrary pages.
  • Not a raw DB search — you do not write SQL or guess column names; you ask a question and the AI routes it.
  • Not your first-party data — it is the market around your business, complementary to your own tables.

​
Where to go next

START HERE — What can you ask?

The intent library: every kind of question the Vertical can answer, grouped by goal, plus the mandatory first-use bootstrap.

AI Operating Rules

The ten grounding rules that keep every answer honest and auditable.

Tool Reference

Each canonical vertical_* tool: purpose, when to call, key inputs, notes.

Business Question Recipes

Natural questions mapped to tool-and-signal flows and answer framing.
For the complete tool list and parameters see Vertical (market intelligence). For how the Vertical fits the wider session contract see the Operating Contract.