An MCP server that enables direct SQL query execution against Microsoft SQL Server databases, integrating with Claude for natural language SQL interactions
This MCP server has critical definition quality gaps. Only 1 tool (execute_sql) is defined, and while it has a description, the input schema is minimal with no parameter descriptions. The tool description is generic and does not explain WHEN to use it, what it returns, or error handling. The schema declares 'query' as a string with a basic description, but lacks critical information about query validation, execution constraints, timeout behavior, or output structure. No error handling documentation exists. The tool appears READ_ONLY based on the risk label, but this is not reflected in the description. The source code shown is entirely a bash installation script (install_sql_mcp.sh) with no visible Python implementation of the actual MCP server or tool registration logic, making it impossible to verify schema completeness, output documentation, or error handling beyond what's stated in the TOOLS metadata.
Execute a SQL query against the configured Microsoft SQL Server database and return results
No output schema documented. The tool description does not specify what fields the response contains, what data types are returned, or how pagination/limits are handled for large result sets.
Input parameter 'query' lacks actionable constraints and validation guidance. No mention of query size limits, timeout behavior, allowed SQL operations, injection protections, or what constitutes a valid vs invalid query.
Tool description is generic and does not explain WHEN to use this tool or how it differs from alternatives. No mention of result limits, pagination, or what to do if a query fails. Does not state whether the tool is read-only or if it can execute writes/deletes.
No error handling guidance. The tool provides no documentation on what errors can occur (syntax errors, timeout, permission denied, etc.), how the LLM should recover, or whether errors are retryable.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 36 | 2026-07-28+ | v2 |
No evidence of tool implementation in provided source code. The only source file shown is a bash installation script; the actual Python MCP server code that implements the execute_sql tool is not visible, making verification of schema, output structure, and error handling impossible.
Parameter description for 'query' is minimal (13 words). According to rubric baselines, parameter descriptions should average ~72 characters and include format, constraints, and examples. This description provides almost no guidance.
No pagination or result limit handling documented. If a query returns thousands of rows, the tool's behavior is undefined, does it truncate? Stream? Block? LLMs need explicit guidance on limits and how to paginate.