Skip to main content
Isthmus exposes three MCP tools that AI models call to explore and query your PostgreSQL database. You don’t call these tools directly — your AI client (Claude, Cursor, etc.) invokes them automatically based on your questions. When an AI model connects to your database for the first time, the most effective sequence is: 
1. discover          → What schemas and tables exist? (sizes, row counts, columns)
2. describe_table    → Full table analysis: columns, types, keys, stats, sample rows, index usage
3. query             → Execute a read-only SQL query (with optional EXPLAIN plan)
Most AI models follow this pattern naturally — they start broad with discover, narrow down to specific tables with describe_table, then write and execute SQL with query.

​
Tool summary

ToolPurposeParameters
discoverFull database landscape: schemas with nested tables(none)
describe_tableComplete table analysis: columns, statistics, sample rows, index usagetable_name (required), schema
queryExecute read-only SQL with optional EXPLAIN planssql (required), explain, analyze

​
Safety guardrails

All tools operate within Isthmus’s safety model:
  • Read-only transactions — queries cannot modify data
  • SQL validation — only SELECT and EXPLAIN statements are allowed (enforced via PostgreSQL AST parsing)
  • Row limits — configurable max rows per query (default: 100)
  • Query timeout — configurable timeout (default: 10s)
  • Schema filtering — only expose the schemas you choose
See Security for the full safety model.