Auto-context injection: get_briefing tool for relevant memories

[arreyder]

Problem

solr-mem requires explicit queries to retrieve memories. Unlike file-based memory systems that auto-load context, agents must know to search and what to search for. This means relevant context is often missed at conversation start.

Proposal

Add a get_briefing MCP tool that returns a curated set of relevant memories based on contextual signals:

get_briefing(
  context: "dynamodb cost optimization",  // free text or keywords
  project: "ops",                          // optional project scope
  max_tokens: 2000,                        // budget for response
  since: "2026-03-25T00:00:00Z"           // optional recency filter
)

Returns the top N most relevant memories, ranked by a combination of:

• Text relevance to context
• Recency
• Importance score
• Whether the memory has been updated since last briefing

Why

In a session today, a dynamo-audit agent stored 19 detailed findings. A separate agent (me) had to explicitly search for them, re-fetch everything on each check, and manually determine what was new. A briefing tool would have surfaced the right context automatically.

Alternatives

• MCP resources with URI templates (e.g., memory://briefing/{topic}) that clients can auto-attach
• A context resource that returns memories matching the current conversation's detected topics