Yes. They are tools. So my approach, at least try to approach is to keep on polishing the skills and check the output of LLM in loops with mcp to alert the abnormality asap so the LLM won’t go to next step to make things worse.