shine-vector-search

Store and retrieve semantic memories using Qdrant MCP vector database. Enables cross-session knowledge recall.

<objective> Manage semantic vector memory via Qdrant MCP:

Store mode (/shine-vector-search store <topic>):

  1. Collect the key information/findings from the current session.
  2. Generate a semantic embedding via the Qdrant MCP.
  3. Store with metadata: topic, timestamp, source files, confidence.
  4. Confirm storage with ID and summary.

Recall mode (/shine-vector-search recall <query>):

  1. Search Qdrant for semantically similar past entries.
  2. Return top 5 matches with relevance score, timestamp, and source.
  3. Present as a structured table with inline previews.

Requirements:

  • Qdrant MCP must be connected: claude mcp add qdrant --command "npx -y @qdrant/mcp-server-qdrant"
  • Qdrant server running: docker run -p 6333:6333 qdrant/qdrant

If Qdrant is not available, fall back to claude-mem plugin for basic text search across memory files, or grep on ~/.claude/memory/. </objective>

<guardrails> - Never store PII, secrets, or credentials in the vector database. - Always include source attribution when recalling stored knowledge. - Label recalled information with its original storage date — it may be stale. - Fall back gracefully: Qdrant → claude-mem → grep on memory dir → ask user. - Never fabricate recalled data — if no match found, say so. </guardrails>