Data Assets

Data Assets is a three-level hierarchy — dataset, table, column — populated by a data-enrichment pass that runs after a data-shaped MCP server (Postgres, Snowflake, and similar) is discovered. Enrichment creates one dataset per server, named "<server name> Data" by convention, with catalog set to the server's own name — a poor-man's discriminator used because there is not yet a direct foreign key from dataset to server, only this name match. schema_name defaults to public when the source system doesn't distinguish schemas.

Table discovery lists the server's tables and creates a row for each one; column discovery is conditional on the server exposing a describe_table tool — if it doesn't, its tables are cataloged with zero columns rather than failing outright. Both passes are idempotent: existing tables and columns are looked up by name before anything is created, so a five-minute re-discovery cycle doesn't insert duplicate rows on every pass. Where a table's name matches an existing resource on the same server, its resource_id gets stamped or backfilled so the two records stay linked.

Sensitivity tagging happens only at the column level, from the column name alone — no sample-value inspection, no LLM. It is deliberately conservative: a column named email or phone gets tagged, but something like user_id is left untagged because it is usually a surrogate key, and a database CHECK constraint limits the tag set to PII, PHI, GDPR, SOX, and HIPAA. Suffixes that indicate derived metadata rather than the sensitive value itself — _count, _status, _verified, and similar — are excluded even when the base name would otherwise match, and an operator can always refine a column's tags by hand afterward. Tables and datasets never get this automatic tagging; only columns do.

/data-queries goes a step further than browsing this metadata — it lets an operator actually run a SQL query against a connected data source, with query suggestions, saved queries, run history, execution stats, and chart visualization of the result set. It is bound to this capability because it is the surface that inspects the SQL actually reaching data assets, not because it browses the same dataset/table/ column rows the rest of this page does.