A Model Context Protocol server for searching, retrieving, and analyzing PDF documents with advanced full-text search capabilities and markdown conversion
This server has 3 tools with acceptable naming (verb_noun pattern) and READ_ONLY semantics, but significant gaps in schema documentation, parameter descriptions, and error handling. Tool descriptions are moderately detailed (90-180 chars), exceeding the minimum but lacking guidance on when to call each vs alternatives. Input schemas are partially visible but lack parameter descriptions, the rubric requires ALL parameters to have descriptions. Output schemas are not documented. No error handling guidance, no pagination support for search results, and no tool annotations. STDIO transport caps the protocolReadiness score at 50.
Retrieve the full converted markdown content of a specific PDF document
List all available PDF documents in the server with their metadata and page counts
Search through PDF documents using full-text search with TF-IDF scoring and proximity boosting for multi-word queries
Input parameter 'max_results' lacks a description; 'query' description is missing constraint details (e.g., minimum length, character restrictions). The rubric requires ALL parameters to have descriptions.
Output schemas are not documented. The LLM cannot plan downstream actions without knowing what fields search_pdfs, list_pdfs, and get_pdf_content return. Baselines show 100% of A+ tools have documented return types.
search_pdfs does not specify pagination (limit/offset or cursor) or document a result cap. Without pagination, returning many results will blow the context window. Baselines show tools should cap results at 20-50.
No error handling guidance. If a PDF is not found, the tool should return 'Document not found. Available documents: [list]' rather than a bare error. Agents need recovery hints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 43 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 40 | - | v1 |
list_pdfs takes no parameters, but the description does not explain what 'metadata' and 'page counts' it returns or their structure. Output schema is undocumented.
No tool annotations (readOnlyHint, idempotentHint, destructiveHint). While all three are READ_ONLY, explicit annotations would help clients optimize behavior and log appropriately.