Shared memory for AI agents — an MCP server that auto-learns your team's conventions and injects them into every session.
Loom defines 18 tools with complete input schemas and descriptions. Naming is verb-forward (learn, teach, reflect, recall_*) and mostly clear. Descriptions are present but often generic (50-150 chars, below the 194-char baseline). Parameters have types and descriptions, but many lack constraints (enums, ranges). Output schemas are not documented in the source. Error handling guidance is absent. The tool set is ambitious but lacks the polish of production-grade servers, descriptions don't explain WHEN to use each tool vs. alternatives, and no recovery guidance for failures.
Amplify coaching across the team
Export rules in structured formats
Export organization timeline (Glen-style)
Ingest rules from another project
Get statistics
Learn from observation
Passive observation (silent auto-capture)
Generate/share onboarding pack for a role
Output schemas not documented. Tools return data but LLMs cannot predict field names, types, or structure. Downstream tool chaining is error-prone.
Descriptions are generic and do not explain WHEN to use each tool vs. similar ones. E.g., recall_memory vs. recall_relevant both search conventions but differ in strategy, this distinction is not explained.
No error handling guidance. Tools do not document what happens on failure (e.g., rule not found, invalid domain, permission denied) or how to recover.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 56 | 2026-07-28+ | v2 |
Search learned conventions
Auto-recall relevant context for a task (Glen-style)
Reflect on completed work
Mark a rule for permanent retention
Auto-preload context at session start (Glen-style)
Set per-observation access control
Backward-compat PR outcome storage
Capture departing member's knowledge
Teach a rule directly
Query the auditable organization timeline
Parameters lack constraints. 'confidence' accepts 1-10 but no enum or range validation. 'format' accepts 'markdown', 'json', 'compact' but no enum. LLMs will hallucinate invalid values.
succession tool has 7 required parameters but only 3 are always needed (member, role, action). Conditional parameters (title, detail, importance, category, domain) should be optional and documented as action-dependent.