FastMCP-based CSV File Connector for Claude. Allows Claude to access and analyze CSV files from your local filesystem
This CSV analyzer server has adequate tool names and descriptions, but suffers from significant schema and output documentation gaps. All four tools are clearly named with action verbs (read_, get_, query_, get_), and each has a basic description (30-50 chars). However, input schemas lack critical details: parameter types are present but descriptions are minimal or missing context about constraints and formats. Output schemas are completely undocumented, the server returns JSON strings with no declared structure, making it impossible for LLMs to know what fields to expect. Error handling returns JSON errors but with no actionable recovery guidance. The server lacks any tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all tools being read-only. Resource endpoints exist but tools do not reference them coherently. Overall, the server is functional for basic CSV analysis but falls short of production-grade tool composition.
Get detailed information about a CSV file
Get statistical summary of a CSV file or specific column
Query a CSV file using pandas query syntax
Read a CSV file and return its contents
Output schemas completely undocumented. All tools return JSON strings with no declared structure. LLMs cannot know what fields to expect (filename, rows, columns, data, shape, statistics, etc.). This violates the response-shaper pattern and forces LLMs to guess at output structure.
Parameter descriptions lack format constraints and context. 'filename' is described as 'Name of the CSV file to read' but provides no guidance on file path handling, allowed characters, or whether paths with subdirectories are supported. 'query' in query_csv lacks any explanation of pandas query syntax syntax.
Error handling provides no recovery guidance. Errors return JSON with generic messages like 'File {filename} not found' or 'Query error: {str(e)}'. LLMs receive no hint about next steps: should they retry? Call a different tool? Ask the user?
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 45 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 36 | - | v1 |
No tool annotations despite all tools being read-only. The server should declare readOnlyHint: true on all four tools to signal safety and idempotency to the agent.
Missing pagination support. Tools like read_csv and query_csv can return large datasets with no limit or pagination parameters. A CSV with 100,000 rows would dump the entire dataset into the response, exhausting token budgets.
Parameter 'rows' in read_csv accepts Optional[int] with no bounds checking. LLMs could pass negative numbers, zero, or enormous values without constraint. Description should specify: 'Optional number of rows to return (1-10000; default: all)'.