Spreadsheet analysis MCP server with natural language queries powered by DuckDB
Server has 8 tools with clear verb-noun naming (load_file, list_tables, run_sql, etc.). All tools have descriptions (10-100 chars, mostly 50-150 range). Input schemas are present with type definitions for all parameters. However, descriptions are terse and lack LLM-optimized guidance on WHEN to use each tool vs alternatives. Parameter descriptions are minimal (e.g., 'Name of the table to list columns for' is adequate but not rich). Output schemas are inferred from Pydantic model_dump() calls but not explicitly documented in tool definitions. Error handling is basic, no recovery guidance, no categorization of error types. No tool annotations (readOnlyHint, destructiveHint) despite clear risk levels. Missing idempotency guarantees and confirmation patterns for destructive clear_all tool.
Reset the DuckDB database and remove all loaded files.
Return schema information for all tables including sample values.
List columns for a specific table with their types.
List all files loaded into data-lens in this session.
List all DuckDB tables currently available.
Load a spreadsheet file (Excel/CSV/Parquet) into DuckDB.
Return the first N rows of a table.
Destructive tool 'clear_all' lacks confirmation pattern and error recovery guidance. No dry-run option or undo mechanism.
Tool descriptions are terse (50-70 chars) and lack LLM-optimized guidance on WHEN to use each tool vs alternatives. Missing dependency hints (e.g., 'Call get_schema before run_sql').
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels documented in metadata. LLM cannot infer which tools are safe to retry.
Error responses return raw exception strings (e.g., 'error': str(e)) with no recovery guidance or categorization. LLM cannot determine if error is retryable or user-fixable.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |
Execute a SQL query against the current DuckDB database.
Output schemas not explicitly documented in tool definitions. LLM must infer structure from Pydantic model_dump() calls, risking misinterpretation of nested fields.