MCP server for T-SQL stored procedure and function analysis, migration guidance, and mapping strategy recommendations to MyBatis/Java
The server provides 3 tools with basic schema definitions but significant quality gaps. Tool names follow verb-noun convention (analyze_sql, tsql_analyze, health), but descriptions lack the depth required for LLM-driven selection and error recovery. Input schemas are visible and typed (string parameters with descriptions), but output schemas are entirely undocumented, critical for agent planning. Parameter descriptions are present but minimal (10-20 chars), below the 50-200 char baseline for LLM optimization. No error handling guidance, no structured output documentation, no parameter constraints or enums. The server implements basic T-SQL analysis logic but exposes none of the rich intelligence (risk levels, recommendations, strategy patterns) visible in the service code to the tool interface.
Analyze a T-SQL statement for references, transactions, control flow, data changes, and migration impacts.
Return basic service health information.
Analyze a T-SQL statement (alias for analyze_sql compatibility).
Output schemas completely undocumented. No response structure defined for any tool. LLMs cannot plan downstream actions or extract fields when output format is invisible.
Descriptions are 10 - 35 characters, well below the 50 - 200 char baseline. Tool descriptions must state WHAT, WHEN, and WHY to use the tool. Current descriptions (e.g., 'Analyze a T-SQL statement') lack context for selection and recovery guidance.
Parameter descriptions are minimal and lack constraints. 'SQL dialect (default: tsql)' should specify: what dialects are valid (enum), what each means, and what happens if an unsupported dialect is passed. No mention of expected SQL size limits, encoding, or statement types.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2025-06-18+ | v2 |
| 2026-03-09 | D | 50 | 2025-11-25+ | v1 |
Duplicate tools: tsql_analyze is an alias for analyze_sql with identical schema. LLM must reason about why two identical tools exist; recommend removing the alias or clearly documenting the distinction.
No error guidance. If SQL parsing fails, invalid dialect is passed, or service is unavailable, there is no documented recovery path. Error responses must tell the LLM what to do next.
Rich analysis logic (risk_signals, confidence scores, recommended patterns, migration strategies) is computed in tsql_mapping_strategy.py but NOT exposed via tool interface. analyze_sql tool discards this intelligence, LLM cannot see migration difficulty, required Java patterns, or specific risks.