Your MCP endpoint

Your agent calls Contextloop for business context before writing SQL. It keeps using its own database tools, so we never need warehouse access.

Then ask your agent a data question. It should call get_context first. Tip: add "Always call contextloop get_context before writing SQL" to your CLAUDE.md or Cursor rules.

Bootstrap your definitions (5 minutes)

Paste this into Claude Code or Cursor once it's connected to both Contextloop and your database. Your agent drafts the definitions; you approve them in the Definitions tab.

Using your database tools, explore our warehouse (schemas, the most important tables, and a few sample rows). Then use the contextloop propose_definition tool to propose:
- the 10 business metrics we most likely care about (with SQL),
- key business terms (e.g. what counts as a customer / active user),
- table notes (which tables to use or avoid, units such as cents, timezones),
- rules about filters we should always apply (test data, internal accounts, deleted rows).
Ask me about anything ambiguous before proposing it. Keep each definition short and precise.

Team

Teammates you add here get this same workspace (definitions, review queue, accuracy) when they sign in with that email. Their agents use the same MCP URL and token.

About your setup (helps us prioritise)