mcp-feedback-collector 配置

一个现代化的 Model Context Protocol (MCP) 服务器，为AI助手提供交互式用户反馈收集功能。对我来说是一个将一次对话增加 10 倍以上的工具

1、mcp 配置

{
  "mcpServers": {
    "mcp-feedback-collector": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-feedback-collector@latest"
      ],
      "env": {
        "MCP_API_KEY": "kkkjjj",
        "MCP_API_BASE_URL": "https://api.ssopen.top",
        "MCP_DEFAULT_MODEL": "grok-3",
        "MCP_WEB_PORT": "5050",
        "MCP_DIALOG_TIMEOUT": "60000",
        "MCP_ENABLE_IMAGE_TO_TEXT": "true"
      }
    }
  }
}

2、rule：

“Before calling the finish tool, or any other action that signifies the completion of the user’s primary request, you MUST FIRST call mcp-feedback-collector.collect_feedback . Only after the user’s feedback is empty, or if the user explicitly indicates satisfaction, should you then proceed to call the finish tool or consider the request fully closed. If feedback is provided, address it and then call mcp-feedback-collector.collect_feedback again, repeating until feedback is empty.”