Server has 19 tools with mixed quality. Naming is generally verb-based (pg_get_*, pg_create_*, pg_manage_*), but many tools lack visible input schemas in source code. Only 3 tools (pg_get_indexes, pg_create_index, pg_execute_sql) have complete, inspectable schemas. Most tool descriptions are present but generic (10-50 chars), falling below the 50-200 char LLM-optimized baseline. Critical issue: 16 of 19 tools have no visible schema definitions in provided source, only tool names and brief descriptions. Error handling exists (sanitizeErrorMessage, audit events) but recovery guidance is absent from tool descriptions.

• Add complete JSON Schema definitions to all 16 tools lacking visible schemas. Use zod-to-json-schema (already in dependencies) to generate schemas from Zod types, matching the pattern in pg_create_index.
• Expand tool descriptions to 50-200 characters. Template: '[VERB] [WHAT] [WHEN to use] [KEY CONSTRAINT]'. E.g., 'Create a new database index on specified columns. Use to optimize slow queries. Supports concurrent creation to avoid locking.'
• Add parameter descriptions to every input field. For each param, state: type, format/constraints, valid values (enum), and example. E.g., 'schema (string): PostgreSQL schema name, default "public". Must exist before creating objects.'
• Add state-change warnings to WRITE/DESTRUCTIVE tool descriptions. Prefix with 'Modifies database:' or 'Destructive:' so LLMs recognize irreversible operations and plan accordingly.
• Implement confirmation pattern for pg_execute_sql: require a 'confirm' boolean parameter or return a dry-run result first, asking user to confirm before executing. Prevents accidental data loss.