A Model Context Protocol (MCP) server for converting PDF files to text, designed for Cursor IDE integration
Single tool with basic schema and description. Tool name follows verb_noun pattern (pdf_to_text). Description present but generic (74 chars, below baseline of 194 chars). Input schema defined with proper JSON Schema structure, but lacks richness expected of production tools. No output schema documented. No error handling guidance. Parameter 'file_paths' is documented but minimal (23 chars). No enum constraints, format specifications, or validation ranges. Tool accepts file paths as strings without input sanitization hints or constraints on path length/character restrictions. No pagination, batching, or composition patterns evident. Security concern: file path input is not validated against path traversal attacks.
Convert PDF file(s) to text for analysis and understanding
Tool description is too generic and brief (74 chars vs baseline 194 chars). Lacks context on when to use, prerequisites, what is returned, and why to call this over alternatives.
Output schema not documented. LLMs cannot predict what fields to expect in the response (text? pages? metadata?). This blocks downstream composition and forces agents to introspect.
No path validation or sanitization guidance. Tool accepts file_paths as raw strings without documented constraints (length, character restrictions, path traversal prevention). Malicious or malformed paths could cause runtime errors or security issues.
Parameter description lacks format/constraint details. 'Array of PDF file paths' does not specify: path format, length limits, wildcard support, relative vs absolute paths, max array size, or error handling for invalid paths.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Error handling does not guide recovery. Errors like 'File not found' or 'Failed to parse PDF' are returned but with no actionable next steps for the LLM (e.g., suggest file listing, retry with different encoding).
No result limit or pagination hints. Tool processes all pages (max: 0) without documenting memory constraints or timeout expectations. Processing large PDFs (1000+ pages) could exhaust resources or degrade agent reasoning by returning massive text blocks.
Tool combines multiple concerns: file existence check, format validation, PDF parsing, and text extraction. Each failure mode is conflated into a single error. Should provide per-file success/failure for batch operations.