MCP server for SQL Server database operations, connection management, schema inspection, and query execution with multi-instance support
The server defines 10 tools with generally clear purposes and good naming conventions (all start with action verbs: Get, Add, Switch, Execute, Read). Most tool descriptions are present and adequately detailed (average ~80-120 chars). Input schemas are visible and use proper JSON Schema with type definitions. However, there are significant gaps: (1) Output schemas are not documented, the source shows no schema definitions for what these tools return, forcing LLMs to reason about response structure blindly. (2) Error handling descriptions are minimal, tools like ExecuteQueryAsync and ReadQueryAsync mention confirmation requirements but do not explain what errors might occur or how to recover. (3) Parameter descriptions vary in quality; some are thorough (e.g., connectionString in AddConnectionAsync includes an example), while others are terse (e.g., 'Get current database connection info' for GetCurrentDatabase lacks specificity about what fields are returned). (4) Two tools (ExecuteQueryAsync and ReadQueryAsync) have overlapping responsibilities and similar parameter names, creating ambiguity for LLM selection. (5) The 'confirmUnsafeOperation' and 'confirm_unsafe_operation' parameters (inconsistent naming across tools 8 and 9) lack guidance on what happens if omitted, do DELETE operations silently fail or proceed? This is critical for agent safety.
Add a new named SQL Server connection. Connectivity is tested before the connection is registered.
Execute a SQL query on the current database with pagination and metadata (supports SELECT, INSERT, UPDATE; DELETE/TRUNCATE require confirmation)
List all configured SQL Server connections with their status and which one is currently active
Get current database connection info
Get a list of all databases on the SQL Server instance with size and backup information
Check connection health and server info
Get a list of all tables in the current database with size information and filtering options
Execute a SQL query with formatting and parameters (SRS: read_query, supports DML with confirmation for DELETE/TRUNCATE)
No output schemas documented. Tools return JSON responses but LLMs have no visibility into field names, types, or structure. ExecuteQueryAsync, ReadQueryAsync, GetDatabasesAsync, and others lack response schema documentation, forcing LLMs to guess at result shape.
Overlapping tools with inconsistent parameter naming. ExecuteQueryAsync (tool 8) and ReadQueryAsync (tool 9) perform similar operations but use different parameter names: 'confirmUnsafeOperation' vs 'confirm_unsafe_operation', 'maxRows' vs 'max_rows', 'pageSize' vs 'max_rows' (semantic mismatch). LLMs cannot reliably distinguish when to use each tool.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 51 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 36 | - | v1 |
Switch the active SQL Server connection to a different named instance. All subsequent tool calls will use the new active connection.
Switch to a different database on the same server
Incomplete error handling guidance. Tools like ExecuteQueryAsync mention 'DELETE/TRUNCATE require confirmation' but do not explain what error is returned if confirmation is false or if the parameter is omitted. Similarly, AddConnectionAsync says 'Connectivity is tested before the connection is registered' but does not specify which error fields or codes indicate connection failure.
Vague descriptions for read-only tools. GetServerHealthAsync is described as 'Check connection health and server info', what fields are returned? Is this a JSON object with uptime, version, connection_state? GetCurrentDatabase says 'Get current database connection info' with no hint about response structure.
ReadQueryAsync uses inconsistent parameter naming and conventions. Parameters like 'max_rows' use snake_case while the tool is PascalCase (ReadQueryAsync). Semantic mismatch: 'format' accepts 'json | csv | table (HTML)' but documentation is unclear, does 'table' mean HTML? What if LLM passes 'xml' or 'parquet'? No enum constraint.
No pagination or result limiting guidance for discovery tools. GetDatabasesAsync lists all databases with optional filters, but no mention of a cap or pagination limit. If a SQL Server instance has 1000+ databases, the tool could return enormous results, exhausting context.