Three tools with minimal descriptions (all under 20 chars in English, though Chinese originals may be longer). Input schemas present but sparse: list_tables has empty schema ({}), describe-table and read_query have basic string parameters with minimal descriptions. No output schemas documented. No error handling guidance, no parameter constraints (enums, ranges), no idempotency hints. Tool names lack clarity on what data is returned or how to chain results. No security annotations or permission declarations.

• Expand tool descriptions to 50-200 characters. Example for read_query: 'Execute a read-only SQL SELECT query against the database. Returns structured rows with column names and values. Use describe-table first to explore schema. Supports SELECT only; INSERT/UPDATE/DELETE are blocked.'
• Document output schemas for each tool. Example: list_tables returns {tables: [{name: string, rows: number, created: ISO8601}]}. describe-table returns {columns: [{name, type, nullable, key, default}]}. read_query returns {columns: [string], rows: [[any]]}.
• Add parameter descriptions with format/constraint hints. Example for read_query sql param: 'SELECT-only SQL query (max 10000 chars). Joins, subqueries, and aggregates supported. INSERT/UPDATE/DELETE/DROP blocked. Timeout: 30s.'
• Implement input validation and return actionable errors. Example: 'Invalid table name "xyz": contains special chars. Valid names: [a-zA-Z0-9_]. Available tables: users, orders, products.'
• Add tool annotations (readOnlyHint: true) to signal safety to LLMs. Mark describe-table and read_query as idempotent (idempotentHint: true) so agents can retry safely.

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

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