// describe_table, as an agent sees it{ "schema": "sales", "name": "orders", "type": "table", "comment": "One row per order placed", "approxRows": 1250000, "columns": [ { "name": "id", "dataType": "integer", "nullable": false, "primaryKey": true }, { "name": "customer_id", "dataType": "integer", "nullable": false, "primaryKey": false }, { "name": "status", "dataType": "smallint", "nullable": false, "primaryKey": false, "comment": "1 open, 2 shipped, 3 cancelled" } ], "foreignKeys": [ { "columns": ["customer_id"], "referencedTable": "sales.customers", "referencedColumns": ["id"] } ], "referencedBy": [ { "table": "sales.order_lines", "columns": ["order_id"], "referencedColumns": ["id"] } ]}Read-only, three times over
Each query must be a single SELECT, checked with the database’s own rules for quotes and
comments. It runs in a transaction that is always rolled back, read-only where the database
supports it. And a database whose login could write is refused altogether.
Four databases
SQL Server, PostgreSQL, MySQL and Oracle, side by side in one server. Agents see each database’s provider, so they write the right dialect the first time.
Schema context for agents
Foreign keys in both directions, table and column comments, and estimated row counts. Agents take join columns from the schema instead of guessing from names.
Results that fit in a context
Rows come back as compact arrays, capped by count and size. A cut-off result says why, and how to narrow or page the query in that database’s syntax.
Look before you query
explain_query returns the estimated plan without running anything, so an agent can spot a
full scan of a large table first. list_tables searches names and comments.
Every call attributed
Each agent gets its own bearer token, and every call is logged under its name, with the SQL, timing and row count. Refused calls are logged too.
Runs where you need it
stdio for a client that launches it, or Streamable HTTP as a shared service in Docker, under systemd or as a Windows Service, with liveness and readiness checks.