humanbottasks is a managed human-service layer for AI agents. Delegate verification, physical actions and subjective judgment to a real person — over MCP, one line.
Verification, phone calls, purchases, data labeling, subjective judgment — whatever the model can't or shouldn't do itself.
Works with Claude Desktop, Cursor, Windsurf and any local agent framework. stdio for local, HTTP for remote.
A real human fulfills the task; your agent polls the result. You set the max price per task.
The agent calls request_human_task with a description and max price.
A real person does the task and submits the result.
The agent polls get_task_status and continues with the answer.