read_data

Use view=catalog for tables, schema_compact for columns, full for complete records/schema, audit for changed-table validation (one table per request; auditAll=true requests all tables). Focus tableId/recordIds/columnIds; includeVersions supplies edit guards. Page with nextCursor + expectedRevision. Details: /docs/mcp/read_data.

Inspect Structured Data efficiently. Start with view=catalog for table names, IDs, descriptions, record counts, stable table links and inbound referencedBy dependencies; columns are omitted. Use view=schema_compact for column IDs/names/types/required/unique/reference definitions, without verbose defaults/descriptions. These views omit versions by default; includeVersions=true opts in. Full reads return versions by default; includeVersions=false omits them. tableId selects a table; columnIds selects schema details for mutation targets, and recordIds/query focuses records. view=audit groups deterministic draft errors by table/column/message with affected counts and sample record IDs; it does not infer unused tables, normalization, requiredness, or invalid cycles. issuesEvaluated=false means validation was skipped for catalog/compact views. Follow nextCursor with the same filters/revision; catalogs paginate tables, full table reads paginate records, audits check changed tables since their last completed audit plus an explicit tableId, one table per request. Reference target changes also invalidate inbound audits. auditAll=true explicitly requests a full audit. auditTableIds identifies the evaluated table and issueCount counts that table only. Checkpoints survive restarts and save after the table's final issue page. Continue until nextCursor is null even when audit is empty. referencedBy gives source table/column and non-null linked record counts; targetColumnId=null means internal record identity. Reference values resolve a target internal record ID first, then the named key when referenceColumnId is set; IDs take precedence over colliding keys. primaryDisplayColumnId controls readable labels. The returned link can be inserted into a Document using existing Markdown/block operations; IDs and links survive rename/type edits. Use table.metaVersion for update_table.expectedMetaVersion, table.version for table/column deletion, table.schemaVersion for add/reorder columns and add_record, column.version for update_column/set_cells, record.version for delete_record. Reuse returned revisions and versions. telemetry measures server work and payload bytes only, excluding connector/network/model time.