MCP MITM Mem0

Instructions

Implementation Reference

• mcp_mitm_mem0/mcp_server.py:603-679 (handler)

  The primary handler function for the 'suggest_next_actions' tool. It delegates to the reflection_agent's suggest_next_steps method to generate and return a list of personalized action recommendations.

  async def suggest_next_actions(user_id: str | None = None) -> list[str]:
      """
      Generate personalized action recommendations based on conversation patterns and history.
  
      ## AUTONOMOUS SUGGESTION TRIGGERS
  
      ### HIGH Priority (Offer Immediately)
      - User asks: "what now?", "what should I do next?", "what's next?"
      - User seems stuck: "I'm not sure...", "what do you think?", "hmm..."
      - Task completion: "that's done", "finished that", "what else?"
      - Repeated issues: 3+ similar questions or problems
  
      ### MEDIUM Priority (Offer Contextually)
      - After successful problem solving: "since that worked, you might also..."
      - Session start with analysis showing patterns
      - User mentions having free time or looking for projects
  
      ### Proactive Integration Examples
  
      ```python
      # User: "That's done. What should I work on now?"
      # → AUTO: suggest_next_actions()
      # → Present: "Based on our recent work, here are some suggestions..."
  
      # Analysis reveals 3 CORS questions
      # → AUTO: suggest_next_actions()
      # → Proactive: "I notice CORS keeps coming up - want to set up a permanent solution?"
  
      # User: "I'm stuck, not sure what to do next"
      # → AUTO: suggest_next_actions()
      # → Response: "Let me suggest some next steps based on your recent projects..."
      ```
  
      ## Enhanced Suggestion Types (Now with Semantic Analysis)
  
      - **Documentation**: Create guides for frequently asked topics
      - **Learning**: Resources for identified knowledge gaps
      - **Workflow**: Process improvements based on patterns
      - **Next Steps**: Logical progression in current projects
      - **Best Practices**: Improvements based on observed anti-patterns
      - **Tools**: Recommendations for tools that could help
      - **Problem Prevention**: Address recurring issues permanently
      - **Project Continuation**: Resume incomplete work
  
      ## Smart Suggestion Generation
  
      Uses semantic search to find:
      1. **Recurring Issues**: Problems that appear multiple times
      2. **Incomplete Projects**: Work that seems unfinished
      3. **Knowledge Gaps**: Areas with frequent questions
      4. **Success Patterns**: Approaches that worked well
      5. **Tool Opportunities**: Where automation could help
  
      ## Response Integration Guidelines
  
      - **Present contextually**: "Since you just finished X, you might want to..."
      - **Be specific**: Include actual examples from conversation history
      - **Prioritize impact**: Lead with suggestions that solve recurring issues
      - **Make actionable**: Each suggestion should be a clear next step
  
      Args:
          user_id: User ID to analyze (optional, defaults to DEFAULT_USER_ID)
  
      Returns:
          List of actionable suggestion strings (max 10), each being:
          - Specific and concrete with examples from history
          - Based on semantic analysis of conversation patterns
          - Prioritized by potential impact and user needs
          - Practical and immediately implementable
      """
      try:
          suggestions = await reflection_agent.suggest_next_steps(user_id=user_id)
          logger.info("Generated suggestions", count=len(suggestions))
          return suggestions
      except Exception as e:
          logger.error("Suggestion generation failed", error=str(e))
          raise RuntimeError(f"Failed to generate suggestions: {str(e)}") from e

• mcp_mitm_mem0/mcp_server.py:599-602 (registration)

  Registration of the 'suggest_next_actions' tool via the @mcp.tool decorator, specifying name and description.

  @mcp.tool(
      name="suggest_next_actions",
      description="Get personalized recommendations - AUTO-SUGGEST when user seems stuck or asks 'what now?'",
  )

• mcp_mitm_mem0/reflection_agent.py:211-284 (helper)

  Supporting helper method in ReflectionAgent class that implements the core suggestion logic by analyzing conversations, searching memories for issues and projects, and generating actionable next steps.

  async def suggest_next_steps(self, user_id: str | None = None) -> list[str]:
      """Suggest next steps based on conversation history using semantic analysis.
  
      Args:
          user_id: User to analyze (defaults to settings)
  
      Returns:
          List of suggested next steps
      """
      user_id = user_id or settings.default_user_id
  
      try:
          # Analyze recent conversations
          analysis = await self.analyze_recent_conversations(user_id=user_id)
          insights = analysis.get("insights", [])
  
          # Search for repeated issues or incomplete projects
          issue_memories = await memory_service.search_memories(
              query="error problem issue bug failed", user_id=user_id, limit=10
          )
  
          project_memories = await memory_service.search_memories(
              query="implement build create project working on",
              user_id=user_id,
              limit=10,
          )
  
          suggestions = []
  
          # Generate suggestions from pattern analysis
          for insight in insights:
              if insight["type"] == "frequent_questions":
                  examples = insight.get("examples", [])
                  if examples:
                      topic = self._extract_topic_from_questions(examples)
                      suggestions.append(
                          f"Create a personal reference guide for {topic} - you've asked about this multiple times"
                      )
                  else:
                      suggestions.append(
                          "Consider creating a FAQ or documentation for commonly asked questions"
                      )
              elif insight["type"] == "focus_area":
                  area = insight["description"].split("on ")[1].split(" (")[0]
                  suggestions.append(
                      f"Deep dive into {area} with more structured learning resources"
                  )
              elif insight["type"] == "problem_solving_pattern":
                  suggestions.append(
                      "Try breaking down the problem into smaller, testable components"
                  )
  
          # Add suggestions based on semantic searches
          if issue_memories:
              recurring_issues = self._identify_recurring_issues(issue_memories)
              for issue in recurring_issues:
                  suggestions.append(
                      f"Document solution for recurring {issue} - appears multiple times"
                  )
  
          if project_memories:
              incomplete_projects = self._identify_incomplete_projects(
                  project_memories
              )
              for project in incomplete_projects:
                  suggestions.append(
                      f"Consider resuming work on {project} - seems unfinished"
                  )
  
          return suggestions[:10]  # Limit to top 10 suggestions
  
      except Exception as e:
          self._logger.error("Failed to suggest next steps", error=str(e))
          return []