Thin FastMCP wrapper around the MSAM REST API. Exposes 10 cognitive memory tools (store, query, context, feedback, graph, decay, contradictions, forget, stats, consolidate) to any MCP-compatible agent via stdio transport.
MSAM MCP server has 10 tools with complete input schemas and descriptions. Naming follows verb_noun convention (msam_store, msam_query, msam_forget). Descriptions are adequate (80-200 chars) and explain WHAT each tool does. However, parameter descriptions lack actionable constraints (no enums for memory_type despite listing valid values; no min/max for numeric params like top_k, max_tokens, depth). Output schemas are not documented, LLMs cannot predict response structure for chaining. Error handling is absent from visible code. No dry-run confirmation for destructive operations (msam_forget, msam_decay). Security: MSAM_API_KEY is injected via env (good), but no per-tool permission declarations or audit logging visible.
Run sleep-based memory consolidation. Merges related memories, extracts patterns, and strengthens frequently co-activated atoms. Mimics biological sleep consolidation.
Get compressed session startup context from MSAM. Retrieves the most relevant and highly-activated memories for the current session, ranked by activation * relevance.
Find contradictory memories in MSAM. Detects memories that conflict with each other, helping maintain consistency in the knowledge base.
Run a decay cycle on MSAM memories. Low-activation memories are demoted or archived. Use dry_run=True to preview what would be affected before committing.
Provide feedback on a memory atom to reinforce or weaken it. Call this after using a memory — positive feedback boosts activation (slower decay), negative feedback accelerates decay.
Output schemas not documented. LLMs cannot predict response structure (fields, types, pagination). Forces agents to guess what data is returned and how to chain tools.
Enum constraints not formalized in schema. memory_type description lists 'episodic, semantic, procedural, working' as text, not as JSON Schema enum. LLMs may hallucinate invalid values.
Numeric parameters lack min/max bounds. top_k, max_tokens, depth, importance have no documented ranges. LLMs may pass absurd values (top_k=999999, depth=100).
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Intentionally forget a memory atom or matching memories.
Explore entity relationships in MSAM's knowledge graph.
Query MSAM for relevant memories using semantic search + activation scoring.
Get MSAM database statistics — atom counts, decay rates, health metrics.
Store a memory atom in MSAM with activation-based decay.
Destructive operations (msam_forget, msam_decay, msam_consolidate) lack dry-run confirmation or explicit warning in description. Agents may irreversibly delete memories without safeguards.
No error handling guidance visible in code. Tool implementations call _post() but no recovery hints, retryability classification, or actionable error messages documented.