This server has 2 tools with partially visible schemas and minimal documentation. Tool names are reasonably verb-based (analyze_*), but descriptions are short and lack context for LLM selection. Input schemas are present but sparse. No output schemas documented. Error handling is basic. The code shows database integration but the tool definitions lack the rigor expected of production-grade agent tools.

• Document the output schema for both tools explicitly. Example: 'Returns {status: string (success|error), message: string, file_path?: string, schema_content?: string}'. This enables LLM to plan downstream actions.
• Expand tool descriptions to 100-150 chars each. For analyze_database_schema: 'Analyzes the structure of a MSSQL database schema (tables, columns, constraints) and exports a human-readable Markdown documentation file. Use this to understand database layout before querying stored procedures. Requires active database connection.'
• For analyze_stored_procedure, clarify: 'Retrieves and explains the SQL definition of a stored procedure in a format optimized for LLM comprehension. Use this to understand what a procedure does before invoking it via API. Procedure name should be fully qualified (e.g., dbo.procedure_name); if not found, returns error with available procedure names.'
• Add parameter constraints: schema_name should document valid schemas or state 'Common values: dbo, public, custom. Other schemas in the database may also be queried.' procedure_name should state 'Fully qualified name preferred (e.g., dbo.ps_test). If schema omitted, defaults to dbo.'
• Implement structured error responses. Instead of returning None on procedure not found, return {status: 'error', reason: 'procedure_not_found', message: 'Procedure dbo.ps_test not found. Available procedures: dbo.sp_users, dbo.sp_orders, ...}. This enables the LLM to offer corrections.
• Add a discovery tool: list_database_schemas() and list_stored_procedures(schema_name) to help LLMs explore the database before calling analyze_* tools. This reduces failed lookups and improves context.