Natural language database explorer via Model Context Protocol
DB Explorer has clear, action-verb tool names (list_, describe_, query, execute_, get_) and all tools include descriptions. However, parameter descriptions are minimal or absent in several tools, and output schemas are not formally documented. The query tool's description is strong (explains translation, validation, execution), but others lack depth. No error handling guidance, no parameter constraints (enums, ranges), and no documentation of output structure. Schemas are inferred from code but not explicitly visible in tool registration.
Describe a table's columns, constraints, and indexes.
Execute a raw SQL SELECT query against the database.
Get sample rows from a table.
List all database tables with approximate row counts.
Ask a natural language question about the database and get results. Translates the question to SQL using Claude, validates the query, executes it, and returns the SQL along with formatted results.
Output schemas not documented. Tools return formatted text strings; LLMs cannot parse structured results or extract fields for downstream calls. No documentation of return format, field names, or data types.
Parameter descriptions missing or minimal. 'table_name' and 'sql' parameters lack guidance on format, constraints, or valid values. LLMs cannot infer whether table_name accepts schema-qualified names or if sql must be SELECT-only.
No error handling guidance. Tools do not document what errors are possible (invalid table, SQL syntax error, timeout) or how LLMs should recover. A timeout or permission error returns no actionable next step.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
No parameter constraints. 'limit' in get_sample_data has no min/max bounds; 'sql' in execute_sql has no validation hint. LLMs may pass absurd values (limit=999999) or non-SELECT queries.
query tool accepts natural language but does not document that it calls Claude internally. No disclosure of external API dependency, latency, or cost implications. Agents cannot reason about when to use query vs execute_sql.