An MCP server for indexing and searching scientific documents using vector embeddings and semantic similarity search
This server has significant definition quality gaps. Tool names lack clear action verbs (memorize_text vs store_text, index_multiple_pdfs vs index_pdfs). Descriptions are present but minimal (avg ~60 chars, well below the 194-char baseline). Input schemas are visible but lack proper type annotations for complex parameters (e.g., 'ctx' parameter in index_multiple_pdfs is typed as 'Context' without description; 'file_paths' and 'texts' are 'list' without element type specs). No output schemas documented. Error handling is minimal, most tools return plain strings without guidance on recovery. The clear_knowledge_base tool is destructive but lacks confirmation/dry-run support. Parameter descriptions are sparse; metadata dict in memorize_text has no guidance on expected keys. No pagination support despite search_knowledge returning variable results. Tool composition is reasonable (single responsibility), but parameter naming could be more explicit (e.g., 'n_results' should be 'max_results' for clarity).
BORRA todo el contenido de la base de datos global. Usar con precaución.
Indexa una lista de archivos PDF (Máx 10 por tanda).
Lista todos los archivos únicos que han sido indexados en la base global.
Memoriza una lista de fragmentos de texto en una sola operación.
Memoriza un fragmento de texto arbitrario en la base global.
Busca en la base de conocimientos científca global.
Input schemas lack proper type definitions. 'file_paths' and 'texts' parameters are typed as 'list' without element type specs (should be 'list of strings'). 'ctx' parameter in index_multiple_pdfs is typed as 'Context' with no description of what it is or why it's needed.
No output schemas documented. Tools return plain strings (e.g., 'Texto memorizado con éxito.') without structured fields. LLMs cannot extract data for downstream tool calls or understand result structure.
Destructive tool (clear_knowledge_base) lacks confirmation or dry-run support. No recovery guidance if called accidentally. Description says 'Usar con precaución' but provides no mechanism to prevent irreversible data loss.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 47 | 2026-07-28+ | v2 |
Parameter descriptions are minimal or missing. 'metadata' in memorize_text has no guidance on expected keys. 'n_results' in search_knowledge lacks range constraints (should specify 1 - 100 or similar). 'file_paths' has no validation rules (max file size, format, path traversal guards).
Error handling is non-actionable. Tools return generic strings like 'Error: No se pudo generar embedding.' or 'Base vacía.' without guidance on what the LLM should do next (retry? call a different tool? ask the user?).