Single tool with acceptable naming but severe deficiencies in parameter description, output schema documentation, and error handling. The tool 'execute_select_query' starts with a verb (positive), but the input schema lacks parameter-level descriptions beyond the query field itself. No output schema is documented, callers must infer the tab-separated format from implementation. Error handling returns a generic TextContent with a string message rather than structured error guidance. The server lacks meaningful composition (only one tool) and provides no pagination, limit controls, or result-shape documentation. Baseline: average tool description length is 194 chars; this one is 56 chars. Parameter descriptions average 72 chars; 'query' has none in the schema (only 'The SELECT query to execute' in the tool description, not in the parameter itself).

• Add a 'description' field to the 'query' parameter in the inputSchema: {"type": "string", "description": "A SELECT query. Only SELECT statements are allowed; UPDATE, DELETE, INSERT, and DDL are rejected. Use table names from the resources endpoint. Results are limited to 10,000 rows; use LIMIT to paginate."}
• Document output schema: 'Returns tab-separated text with column names as the first row, followed by data rows. Example: "id\tname\nemail\n1\tAlice\[email protected]\n2\tBob\[email protected]"'
• Add optional 'limit' and 'offset' parameters to control result size and enable pagination: 'limit (integer, 1-10000, default 100): Maximum rows to return. offset (integer, default 0): Number of rows to skip for pagination.'
• Expand tool description to ~150 - 200 chars: 'Execute a SELECT query on the ClickHouse database. Only SELECT queries are allowed; UPDATE/DELETE/INSERT/DDL are rejected. Useful for retrieving data from tables listed in the resources endpoint. Results are limited to 10,000 rows; use LIMIT to paginate large datasets. Queries timeout after 30 seconds.'
• Replace generic error handling with recovery guidance: Instead of 'Error executing query: <raw error>', return 'Query failed: <parsed error detail>. Check table/column names in the resources endpoint. If the query timed out, use a smaller LIMIT or add WHERE clauses to filter data.'
• Add a 'tables' discovery parameter or a separate 'list_tables' tool so the LLM can discover available tables and columns without relying solely on resources (which list full schema but not dynamically).

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

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