Server has 5 tools with clear verb-noun naming (list_*, describe_*, query, query_log). All tools have descriptions (avg ~150 chars, within baseline 34-392). Input schemas are present with typed parameters and descriptions. However, output schemas are undocumented, tools return formatted strings rather than structured objects, forcing LLMs to parse unstructured text. Error handling is minimal: no recovery guidance, no categorization of retryable vs fatal errors. Security is strong (credentials via keychain, audit logging, permission checks), but output lacks pagination support and result limits are not exposed in descriptions.

• Return structured JSON objects instead of formatted strings. E.g., query() should return {status, query_type, rows, columns, rows_affected, execution_ms, error?} so LLMs can extract and chain data.
• Add tool annotations via FastMCP: @mcp.tool(annotations={readOnlyHint: true}) for list_* and describe_*, destructiveHint: true for query() when it detects INSERT/UPDATE/DELETE.
• Enhance error messages with recovery guidance. E.g., 'Connection "prod" not found. Available: ' + list(config.connections.keys()). Or 'Query denied: requires write permission. Current: read. Contact admin or use a different connection.'
• Document result limits and pagination in descriptions. E.g., 'Returns up to 20 most recent entries. Use limit parameter (1-100) to adjust. No cursor-based pagination; use connection filter to narrow results.'
• Add parameter dependency hints. E.g., connection param: 'Name of a configured connection (call list_connections to see available options).'