Happy slice of mcp pizzaMCP.pizza

qzcli_tool

MCP.Pizza Chef: tianyilt

qzcli is a command-line tool for the Qizhi computing platform, and it ships a connector so Claude Code or Codex can use it directly. From a chat you can sign in, see how many machines each compute group has spare, list your running jobs, submit a new one, and read a job's logs without switching windows. The written guide is in Chinese, and it only works if you already have an account on that platform. Installing it takes a couple of terminal commands.

Coding
Data

Use This MCP server To

Check how many GPU nodes are free right now See which of my training jobs are still running Read the logs from a job without leaving the chat Find a compute group with four free nodes Sign in without copying login cookies by hand Submit a training job without remembering every flag

README

qzcli - 启智平台任务管理 CLI

Release Python Tests License

一个类似 kubectl / docker 风格的启智平台命令行工具，把资源查询、任务提交、任务管理、日志查看和 MCP/agent 工作流收敛到 CLI 里完成。

项目链接

特性

  • 一键登录: qzcli login 通过 CAS 认证自动获取 cookie，无需手动复制
  • 资源发现: qzcli res -u 调 cluster_info / task_dimension 聚合工作空间、计算组、项目并本地缓存（默认 quick 模式秒级返回；需要更新 specs 时加 --full 走全量历史任务扫描）
  • 节点查询: qzcli avail 查询各计算组空余节点，支持低优任务统计和 cookie 失效自动刷新
  • 可视化看板: qzcli dashboard 用 treemap 按「计算组→优先级→项目→用户→任务」逐层下钻 GPU 占用（需 pip install 'qzcli[dashboard]'）
  • 交互式提交: qzcli create -i 提供层级式选择界面，缺少快照时按需预加载
  • 任务列表: 美观的卡片式显示，完整 URL 方便点击
  • 日志查看: qzcli logs <job-id> 直连平台日志接口，支持 tail、follow、raw/json 输出
  • 状态监控: watch 模式实时跟踪任务进度

快速查看资源：

qzcli login -u 用户名 -p 密码 && qzcli avail

qzcli_tool FAQ

Which apps does this work in?
Claude Code and Codex. Setup is a single add command in either one.
Do I need an account?
Yes. It only works if you already have login details for the Qizhi platform. It signs in with those and keeps the session for you.
Can I use this to check for free GPUs before I submit a job?
Yes — that is one of its main jobs. Ask for free machines overall, or narrow it to a single workspace or compute group.
How hard is setup?
You need to be comfortable in a terminal. Copy the project down, install it with pip, then run one command to register it with your assistant.
What language is the documentation in?
Chinese. The commands themselves are English, but the guide and many of the output labels are not.
Is there anything to look at besides the chat?
Yes. An optional dashboard draws GPU usage as a clickable map broken down by compute group, priority, project, user, and job.