About this guide

The documentation for people who use and run semantic-warehouse-mcp. This folder is the single source: the site (Astro + Starlight, in website/, at swm.pipelab.dev) is built from these files and carries every one of them — its build stops otherwise; edit them here, never in the site.

How the project is built and planned is not here: that is dev/ (design, plan, decisions, research).

Pages

• Getting started — from nothing to an agent answering about a made-up shop’s data.
• Reference
  • Tools — every tool, its inputs and answers.
  • Configuration — every key of the configuration file, and its check.
  • Semantic models and vocabularies — Apache Ossie documents, your own metadata declared in a vocabulary, every check at start.
  • The dbt importer — a dbt project’s tables, descriptions, meta and tags as an Ossie document, and its configuration.
  • The guard — how a query is read, every refusal and why, the limits.
  • Engines — the databases it reads, and how each is kept read-only.
  • Running it — the command, stdio or HTTP for a team (its token, the health check), the container image.

Planned, each written with the change that makes it true:

• Guides — connecting databases (read-only users, readable, large tables, secrets by name); writing a semantic model step by step (the reference is semantic models and vocabularies); context documents and skills; saved query sets; running it for a team.
• Contributing — how the project is built and tested.

Three readers: people who use it through their agent (getting started, guides), people who run it (guides, reference), people who contribute (a contributing page). A page is written in the same commit as the change that makes it true; a tool without its reference entry fails the tests, and so does a refusal, a configuration key or an engine without its entry.