MCP server for managing assisted user routines, caregiver alerts, and audit events with safety policy enforcement and role-based access control
MemoryBridge MCP Routines Server has 11 tools with complete input schemas and descriptions. Naming follows verb_noun convention (get_, create_, update_, mark_, approve_, reject_, list_). Descriptions are present (avg ~100 chars) but lack actionable context for LLM selection. Parameters are typed and mostly described, but lack constraints (enums, ranges, patterns). Output schemas are not documented. Error handling and recovery guidance are absent. Security-critical operations (approve_routine, create_routine_draft) lack confirmation/dry-run patterns. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear read/write distinction.
Approve a routine draft. Enforces caregiver authorization.
Create a caregiver alert for help requests or missed routines.
Create a new routine in draft state. Performs structural safety policy checks.
Retrieve approved contacts list for an assisted user.
Retrieve a single routine by ID. Enforces caregiver/user authorization.
Retrieve today's active routines for a user.
No output schemas documented. LLMs cannot infer response structure, forcing them to guess what fields are returned and what downstream tools accept. This breaks tool chaining and wastes tokens on clarification.
Destructive/approval operations (approve_routine, reject_routine, create_routine_draft with safety_decision) lack confirmation or dry-run patterns. Agents can irreversibly approve unsafe routines without a confirmation step.
Enum constraints missing for status/risk_level parameters. mark_routine_status accepts enum ['completed','help_requested','missed'] but create_routine_draft risk_level is free-form string. LLMs will hallucinate invalid values.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 57 | <=2025-11-25 | v2 |
Retrieve approved preferences for an assisted user.
List routines belonging to the caregiver's approved assisted users.
Mark a routine's status (completed, help_requested, missed).
Reject an unapproved routine draft. Enforces caregiver authorization.
Update routine details. Reruns deterministic safety checks and resets approval status on content changes.
No error handling guidance. Tools do not document what errors are retryable, user-fixable, or fatal. LLMs cannot determine recovery strategy when calls fail.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Risk field in tool metadata indicates read vs write, but MCP annotations are absent. Agents cannot infer safety properties from schema alone.