We standardized on @ClickUp for task management and agentic AI. Then we connected our Software Factory to their hosted MCP server and found at least 27 significant bugs. Task assignees silently disappear. “1.5h” gets logged as 1 minute. Comments lose text. Date filters omit results. Updates report success without changing anything. Some calls return an error after making the change anyway. And so on... For AI agents, false success is especially damaging: the agent moves on believing the work is done, while the underlying task is wrong. We’ve reported every issue to ClickUp Support. If you’re evaluating ClickUp for AI-agent integration, test the actual results carefully. Right now, we can’t trust their hosted MCP server’s responses.