A personal memory database MCP server with BM25 full-text search and temporal reranking
Three tools with clear verb-noun naming (ingest_document, search_memory, scrub_document) and documented input schemas. Descriptions are present but brief (18-60 chars, below the 50-200 char LLM-optimized baseline). All parameters have type definitions and descriptions. Output schema is not explicitly documented, search_memory returns RankedMemoryRow but the response format is not visible in tool registration. Error handling is present but generic (FILE_READ_ERROR_PREFIX). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk profiles. Missing output schema documentation is a significant gap for agent planning.
Ingest a document into the personal memory database.
Remove ingested document entries from the memory database by filename.
Search memory using BM25 with temporal reranking.
Output schemas not documented. search_memory returns RankedMemoryRow (with ageInDays, finalScore, bm25_score fields) but tool registration does not declare this structure. Agents cannot plan downstream operations or extract fields reliably.
Tool descriptions are too brief (18-60 chars vs. 50-200 char baseline). 'Ingest a document into the personal memory database' lacks context on when to use it, prerequisites, or side effects. 'Remove ingested document entries from the memory database by filename' does not warn that this is destructive.
No tool annotations despite clear risk profiles. scrub_document is destructive (deletes records) but lacks destructiveHint. search_memory is read-only but lacks readOnlyHint. ingest_document modifies state but lacks annotation. Agents cannot reason about safety without explicit hints.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 52 | 2026-07-28+ | v2 |
Error handling is generic. ingest_document returns FILE_READ_ERROR_PREFIX + message, but does not guide recovery (e.g., 'File not found. Check the path and try again.'). scrub_document silently returns 0 if no documents match, agent cannot distinguish 'file not found' from 'file exists but no entries'.
search_memory limit parameter defaults to 5 but description does not explain why or when to override. No guidance on pagination or result truncation. Agents may not realize they are seeing only a subset of matches.