brain-mcp

Instructions

    Track the evolution of thinking about a topic over time.

    Args:
        topic: The concept/term to track
        view: What to show:
            - "full" (default): Complete trajectory with genesis, temporal pattern, semantic matches, thinking stages
            - "velocity": How often the concept appears over time with trend analysis
            - "first": When the concept first appeared — the genesis moment
    

Implementation Reference

• brain_mcp/server/tools_synthesis.py:312-375 (handler)

  The `thinking_trajectory` tool is defined here as an MCP tool, which tracks the evolution of thoughts on a given topic over time.

  @mcp.tool()
  def thinking_trajectory(topic: str, view: str = "full") -> str:
      """
      Track the evolution of thinking about a topic over time.
  
      Args:
          topic: The concept/term to track
          view: What to show:
              - "full" (default): Complete trajectory with genesis, temporal pattern, semantic matches, thinking stages
              - "velocity": How often the concept appears over time with trend analysis
              - "first": When the concept first appeared — the genesis moment
      """
      if view == "velocity":
          return _concept_velocity(topic)
      elif view == "first":
          return _first_mention(topic)
  
      output = [f"## Thinking Trajectory: '{topic}'\n"]
      cfg = get_config()
  
      # 1. Semantic matches from embeddings
      embedding = get_embedding(topic)
      semantic_results = []
      if embedding and cfg.lance_path.exists():
          lance_results = lance_search(embedding, limit=20, min_sim=0.3)
          semantic_results = [(r[2], r[3], r[0], r[1], r[4]) for r in lance_results]
  
      # 2. Keyword matches with temporal distribution
      try:
          conv_con = get_conversations()
          pattern = f"%{topic}%"
          temporal_dist = conv_con.execute("""
              SELECT
                  strftime(created, '%Y-%m') as period,
                  COUNT(*) as mentions
              FROM conversations
              WHERE content ILIKE ? AND role = 'user'
              GROUP BY period
              ORDER BY period
          """, [pattern]).fetchall()
      except Exception:
          temporal_dist = []
  
      # 3. First mention
      try:
          first_mention = conv_con.execute("""
              SELECT created, conversation_title, substr(content, 1, 200) as preview
              FROM conversations
              WHERE content ILIKE ? AND role = 'user'
              ORDER BY created ASC
              LIMIT 1
          """, [pattern]).fetchone()
      except Exception:
          first_mention = None
  
      if first_mention:
          output.append("### Genesis")
          output.append(f"**First appeared**: {str(first_mention[0])[:10]}")
          output.append(f"**Context**: {first_mention[1] or 'Untitled'}")
          output.append(f"> {first_mention[2]}...\n")
  
      if temporal_dist:
          output.append("### Temporal Pattern")
          total = sum(t[1] for t in temporal_dist)