Treasure Data MCP Server

Instructions

Smart analysis tool that accepts URLs from error alerts or IDs. Automatically
detects type and provides comprehensive execution analysis with recommendations.

Accepts:
- Console URLs (e.g., https://console.../app/sessions/123456)
- Session IDs (e.g., 123456789)
- Attempt IDs (e.g., 987654321)

Common scenarios:
- Someone shares a workflow URL in Slack during incident
- Quick analysis when you only have an ID from logs
- One-stop debugging for any execution issue

Returns analysis with failures, slow tasks, and actionable recommendations.

Implementation Reference

• td_mcp_server/execution_tools.py:318-484 (handler)

  The td_analyze_execution tool handler function. Analyzes a workflow execution given a console URL or session/attempt ID. Extracts IDs from URLs, fetches session or attempt details, retrieves tasks, identifies failures and long-running tasks, and generates recommendations.

  async def td_analyze_execution(url_or_id: str) -> dict[str, Any]:
      """Analyze workflow execution from console URL or ID - best for debugging.
  
      Smart analysis tool that accepts URLs from error alerts or IDs. Automatically
      detects type and provides comprehensive execution analysis with recommendations.
  
      Accepts:
      - Console URLs (e.g., https://console.../app/sessions/123456)
      - Session IDs (e.g., 123456789)
      - Attempt IDs (e.g., 987654321)
  
      Common scenarios:
      - Someone shares a workflow URL in Slack during incident
      - Quick analysis when you only have an ID from logs
      - One-stop debugging for any execution issue
  
      Returns analysis with failures, slow tasks, and actionable recommendations.
      """
      if not url_or_id or not url_or_id.strip():
          return _format_error_response("URL or ID cannot be empty")
  
      import re
  
      # Try to extract IDs from URLs
      session_id = None
      attempt_id = None
  
      # Check for session URL
      session_match = re.search(r"/sessions/(\d+)", url_or_id)
      if session_match:
          session_id = session_match.group(1)
      # Check for attempt URL
      attempt_match = re.search(r"/attempts/(\d+)", url_or_id)
      if attempt_match:
          attempt_id = attempt_match.group(1)
      # If no URL pattern, assume it's a direct ID
      elif url_or_id.isdigit():
          # Try to determine if it's a session or attempt by checking both
          session_id = url_or_id  # Will try session first
  
      client = _create_client(include_workflow=True)
      if isinstance(client, dict):
          return client
  
      try:
          analysis = {"input": url_or_id}
  
          # Try session first
          if session_id:
              session = client.get_session(session_id)
              if session:
                  analysis["type"] = "session"
                  analysis["session"] = {
                      "id": session.id,
                      "workflow": session.workflow["name"],
                      "project": session.project["name"],
                      "session_time": session.session_time,
                      "status": session.last_attempt.status,
                      "success": session.last_attempt.success,
                  }
  
                  # Get attempt details
                  attempt_id = session.last_attempt.id
                  analysis["latest_attempt_id"] = attempt_id
  
          # If not a session or if we have an attempt ID, try attempt
          if attempt_id and "type" not in analysis:
              attempt = client.get_attempt(attempt_id)
              if attempt:
                  analysis["type"] = "attempt"
                  analysis["attempt"] = {
                      "id": attempt.id,
                      "workflow": attempt.workflow["name"],
                      "project": attempt.project["name"],
                      "session_time": attempt.session_time,
                      "status": attempt.status,
                      "success": attempt.success,
                  }
                  attempt_id = attempt.id
  
          # If we found something, get tasks for detailed analysis
          if attempt_id:
              tasks = client.get_attempt_tasks(attempt_id)
  
              # Analyze task execution
              failed_tasks = []
              long_running_tasks = []
              task_tree = {}
  
              for task in tasks:
                  # Build task tree
                  if task.parent_id:
                      if task.parent_id not in task_tree:
                          task_tree[task.parent_id] = []
                      task_tree[task.parent_id].append(task.full_name)
  
                  # Track failures
                  if task.state in ["failed", "error"]:
                      failed_tasks.append(
                          {
                              "name": task.full_name,
                              "state": task.state,
                              "error": task.error if task.error else "No error details",
                          }
                      )
  
                  # Track long-running tasks
                  if task.started_at and task.updated_at:
                      try:
                          from datetime import datetime
  
                          started = datetime.fromisoformat(
                              task.started_at.replace("Z", "+00:00")
                          )
                          updated = datetime.fromisoformat(
                              task.updated_at.replace("Z", "+00:00")
                          )
                          duration = (updated - started).total_seconds()
                          if duration > 300:  # Tasks longer than 5 minutes
                              long_running_tasks.append(
                                  {
                                      "name": task.full_name,
                                      "duration_seconds": duration,
                                  }
                              )
                      except Exception:
                          pass
  
              analysis["task_analysis"] = {
                  "total_tasks": len(tasks),
                  "failed_tasks": failed_tasks,
                  "long_running_tasks": sorted(
                      long_running_tasks,
                      key=lambda x: x["duration_seconds"],
                      reverse=True,
                  )[:5],  # Top 5 longest
              }
  
              # Add recommendations
              recommendations = []
              if failed_tasks:
                  recommendations.append(
                      f"Found {len(failed_tasks)} failed task(s). "
                      "Check error details above."
                  )
              if long_running_tasks:
                  recommendations.append(
                      f"Found {len(long_running_tasks)} task(s) running longer than "
                      "5 minutes. Consider optimization."
                  )
              if not analysis.get("session", {}).get("success", True):
                  recommendations.append(
                      "Workflow failed. Use td_get_attempt_tasks for detailed "
                      "task breakdown."
                  )
  
              if recommendations:
                  analysis["recommendations"] = recommendations
  
              return analysis
          else:
              return _format_error_response(
                  f"Could not find session or attempt with identifier: {url_or_id}"
              )
  
      except Exception as e:
          return _format_error_response(f"Failed to analyze execution: {str(e)}")

• td_mcp_server/execution_tools.py:25-31 (registration)

  Registers the td_analyze_execution tool using the mcp.tool() decorator within the register_execution_tools function.

  # Register all tools
  mcp.tool()(td_get_session)
  mcp.tool()(td_list_sessions)
  mcp.tool()(td_get_attempt)
  mcp.tool()(td_get_attempt_tasks)
  mcp.tool()(td_analyze_execution)

• td_mcp_server/mcp_impl.py:722-723 (registration)

  Invokes register_execution_tools from execution_tools module to register all execution tools, including td_analyze_execution, with the MCP server instance.

  # Register workflow execution tools
  execution_tools.register_execution_tools(mcp, _create_client, _format_error_response)