Self-organizing memory layer for LLMs. A self-hosted MCP server that gives any LLM persistent, self-organizing memory.
MemFabric has 9 tools with clear verb-noun naming (get_rules, list_memories, read_memory, remember, update_memory, delete_memory, start, export_memory, import_memory). All tools have descriptions (10 - 200 chars, within baseline 10 - 1024 range). Input schemas are present and typed for all tools. However, parameter descriptions are minimal or absent in several tools (e.g., 'merge' in import_memory lacks guidance on behavior). Output schemas are not documented, LLMs cannot predict response structure. Error handling is absent; no recovery guidance or categorization. The server lacks tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY, WRITE, DESTRUCTIVE). Composition is sound: tools are single-purpose and chainable. Overall: solid naming and basic schemas, but incomplete descriptions, missing output documentation, and no error guidance.
Delete a memory file
Export all memory files as a zip archive
Get the current memory rules and onboarding status
Import memory files from a zip archive
List all memory files with their entry counts and preview
Read the full content of a memory file
Append a dated entry to a memory file (creates file if it doesn't exist)
Output schemas not documented. LLMs cannot predict response structure (e.g., what fields list_memories returns, what export_memory yields). This forces agents to guess and wastes tokens on trial-and-error.
Tool annotations missing. Despite clear risk levels (READ_ONLY, WRITE, DESTRUCTIVE), tools lack readOnlyHint, destructiveHint, idempotentHint annotations. Agents cannot distinguish safe from dangerous operations without reading descriptions.
Parameter descriptions incomplete. 'merge' in import_memory lacks guidance on behavior (what happens to conflicts? are duplicates merged or replaced?). 'filename' in remember/update_memory/delete_memory lacks format constraints (kebab-case requirement buried in server description, not param docs).
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 64 | <=2025-11-25 | v2 |
Begin onboarding: initialize the memory system and set onboarding status
Replace the entire content of a memory file
No error handling or recovery guidance. Tools fail silently or with generic errors. No guidance for LLM on what to do if a file is not found, if a write fails, or if import_memory encounters a corrupt zip.
Destructive operations lack confirmation step. delete_memory and import_memory (with merge=false) are irreversible but have no dry-run or confirmation mechanism. Agents can accidentally destroy data.