Server has 8 tools with basic schemas and descriptions, but significant quality gaps. All tools have descriptions (10-60 chars, mostly short), and input schemas are present with type definitions. However, descriptions are generic and lack actionable context for LLM selection. No output schemas documented. Error handling returns empty strings on failure with no recovery guidance. No parameter validation constraints (enums, ranges). Tool names are clear (verb_noun pattern), but descriptions don't explain WHEN to use each tool or what distinguishes similar operations (e.g., update_values vs create_table). No tool annotations (readOnlyHint, destructiveHint). Security concern: SQL injection risk, tools accept raw SQL strings without sanitization or validation.

• Document output schemas for all tools. E.g., get_databases returns {columns: string[], rows: [string, ...][]}. Include this in tool descriptions or a separate schema field.
• Add SQL injection protection: use parameterized queries (psycopg prepared statements) instead of string interpolation. Accept structured input (table name, column definitions) rather than raw SQL for create_table and drop_table.
• Enhance error messages. Return structured errors: {error: string, code: string, retryable: boolean, suggestion: string}. E.g., 'Database not found. Available: [db1, db2]. Retryable: false.'
• Add tool annotations: mark query_data, get_databases, get_schemas, get_server_config with readOnlyHint=true; mark drop_table, update_values, create_table with destructiveHint=true.
• Expand descriptions to 50-150 chars with WHEN context. E.g., 'Updates or inserts rows into an existing table using SQL INSERT/UPDATE. Use after create_table. Requires valid SQL syntax.'

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry