MCP Standards

Instructions

Implementation Reference

• src/mcp_standards/intelligence/claudemd_manager.py:336-380 (handler)

  Core handler logic: Queries the tool_preferences table for high-confidence, unvalidated learned preferences matching project/global scope, returns up to 10 suggestions with metadata.

  def suggest_updates(
      self,
      project_path: Optional[str] = None,
      min_confidence: float = 0.7
  ) -> List[Dict[str, Any]]:
      """
      Get suggestions for CLAUDE.md updates without applying them
  
      Returns:
          List of suggested updates with metadata
      """
      suggestions = []
  
      with sqlite3.connect(self.db_path) as conn:
          conn.row_factory = sqlite3.Row
  
          # Find preferences not yet in CLAUDE.md
          if project_path:
              where_clause = "project_path = ? AND project_specific = TRUE"
              params = (project_path, min_confidence)
          else:
              where_clause = "project_specific = FALSE"
              params = (min_confidence,)
  
          cursor = conn.execute(f"""
              SELECT category, preference, confidence, apply_count, learned_from
              FROM tool_preferences
              WHERE {where_clause}
                AND confidence >= ?
                AND last_validated IS NULL
              ORDER BY confidence DESC, apply_count DESC
              LIMIT 10
          """, params)
  
          for row in cursor.fetchall():
              suggestions.append({
                  "category": row["category"],
                  "preference": row["preference"],
                  "confidence": row["confidence"],
                  "apply_count": row["apply_count"],
                  "learned_from": row["learned_from"],
                  "recommended": row["confidence"] >= 0.8
              })
  
      return suggestions

• src/mcp_standards/server.py:579-598 (handler)

  MCP tool wrapper handler: Delegates to ClaudeMdManager.suggest_updates and returns formatted JSON response.

  async def _suggest_claudemd_update(
      self, project_path: Optional[str] = None, min_confidence: float = 0.7
  ) -> Dict[str, Any]:
      """Get suggestions for CLAUDE.md updates"""
      try:
          suggestions = self.claudemd_manager.suggest_updates(
              project_path=project_path,
              min_confidence=min_confidence
          )
  
          return {
              "success": True,
              "suggestions": suggestions,
              "count": len(suggestions),
              "project_path": project_path or "global",
              "message": f"Found {len(suggestions)} suggestion(s) for CLAUDE.md update"
          }
      except Exception as e:
          return {"success": False, "error": str(e)}

• src/mcp_standards/server.py:215-225 (registration)

  MCP tool registration defining name, description, and input schema.

  Tool(
      name="suggest_claudemd_update",
      description="Get suggestions for CLAUDE.md updates based on learned patterns (does not apply them)",
      inputSchema={
          "type": "object",
          "properties": {
              "project_path": {"type": "string", "description": "Project path for project-specific suggestions (optional)"},
              "min_confidence": {"type": "number", "description": "Minimum confidence threshold", "default": 0.7},
          },
      },
  ),

• src/mcp_standards/server.py:292-297 (handler)

  Dispatch logic in the main call_tool handler that routes tool calls to _suggest_claudemd_update.

  elif name == "suggest_claudemd_update":
      result = await self._suggest_claudemd_update(
          arguments.get("project_path"),
          arguments.get("min_confidence", 0.7)
      )
      return [TextContent(type="text", text=json.dumps(result))]

• src/mcp_standards/model_router.py:41-42 (helper)

  Model routing complexity classification for cost-optimized execution.

  "suggest_claudemd_update": TaskComplexity.MODERATE,
  "calculate_significance": TaskComplexity.MODERATE,