crystal_wake

​

What it does

​

Parameters

​

Returns

{
  "briefing": "string",
  "openGoals": [
    {
      "title": "string",
      "memoryId": "string"
    }
  ],
  "recentDecisions": [
    {
      "title": "string",
      "memoryId": "string"
    }
  ]
}

• briefing — Human-readable overview of session context and active priorities
• openGoals — Array of goal memories marked as active/pending
• recentDecisions — Array of recent decision memories (typically from last 7 days)
• Each memory includes title and memoryId for reference/drill-down

• The briefing text
• Open goals (bulleted list, or “none” if empty)
• Recent decisions (bulleted list, or “none” if empty)

​

Examples

​

Basic usage

// Get default session briefing
crystal_wake()

// Text output:
// Briefing: You are working on the Memory Crystal MCP server...
//
// Open goals
// - Implement KB filtering for multi-tenant isolation
// - Add vector search optimizations
//
// Recent decisions
// - Chose Convex mutations over HTTP actions for memory writes
// - Switched from single embedding to cached batch processing

​

Advanced usage

// Get channel-scoped briefing
crystal_wake({
  channel: "claude-code"
})

// Get briefing for a specific project context
crystal_wake({
  channel: "telegram-coaching"
})

// Result includes goals/decisions specific to that channel

​

When to use this vs alternatives

• crystal_wake — At session start to establish context; summarizes goals and recent decisions
• crystal_recall — When you need semantically relevant memories on a specific topic
• crystal_what_do_i_know — When you want a broad topic scan without fixed recency/category filters
• crystal_recent — When you need recent message history, not memory summaries

​

Error handling

​

Notes

• The briefing is generated fresh on each call, not cached.
• Open goals are memories with category=goal and status=active. Recent decisions are category=decision within the last 7 days.
• Channel filtering is optional; if omitted, returns cross-channel goals and decisions.
• The formatted briefing output includes a defense header to prevent accidental instruction injection.
• If no goals or decisions exist, the tool returns empty arrays but does not error.
• The briefing is designed to be read by an AI before substantial work begins; it primes context without forcing action.