Skip to main content
Collections are Driftless’s operational substrate: pipelines, CRM rows, trackers, and other structured work. Their MCP tools are action-based, bounded by default, and use the same server contracts as the dashboard and CLI. Writes require work:write; destructive collection purge additionally requires a human owner/admin identity.

Collections and records

Use driftless_collection_query for bounded reads. Before acting on a record, load the Collection’s criterion with driftless_collection action:'context'; it returns the Knowledge that governs how the team handles those records.

Entities

An Entity is a cross-collection identity that records can reference. driftless_entity action:'upsert' is idempotent on (kind, dedup_key).

Permissions

MCP clients cache tool schemas per connector session. Reconnect the connector after an upgrade so removed or changed tools are reflected in the client.