MCP server for executing SQL queries against MySQL databases
Single tool 'execute_sql' has critical gaps. Description is minimal (23 chars, below 20-char threshold). Input schema present but lacks parameter type validation and constraints. No output schema documented. Tool accepts arbitrary SQL with WRITE risk but provides no error guidance, recovery hints, or input sanitization warnings. No security controls visible for SQL injection prevention. Naming is acceptable (verb_noun), but the tool is dangerously broad and lacks the specificity required for safe agent use.
Execute any SQL query
Description too short (23 chars). 'Execute any SQL query' provides no context on when to use, what it returns, or that it modifies state. LLMs cannot determine safe vs unsafe invocations.
No output schema documented. Tool returns ResponseType (List of TextContent/ImageContent/EmbeddedResource) but LLMs cannot predict structure. Rows are formatted as '[r.cells for r in rows]', unclear if this is a list of dicts, strings, or nested objects.
No input validation or SQL injection protection visible. Tool accepts raw SQL string with no sanitization, parameterization hints, or warnings. Agents can be tricked into passing malicious payloads.
Error handling returns generic 'Error: {str(e)}' with no recovery guidance. LLMs receive raw exception messages (e.g. 'Access denied', 'Syntax error') with no hint on what to do next or how to self-correct.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 34 | 2026-07-28+ | v2 |
Tool marked WRITE risk but no confirmation/dry-run pattern. Agents can execute destructive queries (DROP TABLE, DELETE) without safeguards. No idempotency guarantee or undo mechanism.