Mcpet

MCP desktop companion for AI agent status, usage tracking, and everyday tools. Anything Universe · 桌面伙伴与日常工具入口。

LLM Mart 4 views 77k listing impressions

A desktop pet that watches your agent sessions, your local token usage, and your lunch.

一只桌宠:盯着 Agent 会话、本机用量,还有你的饭点。

Release Python Platform

DeepSeek 娘待机动画 / DeepSeek pet idle animation DeepSeek 娘忙碌工作动画 / DeepSeek pet working animation DeepSeek 娘麦麦套装待机动画 / McDonalds outfit idle animation

GIFs shipped in this repository under docs/media/deepseek/; sources and licenses are listed in docs/media/deepseek/SOURCE.md.
动画文件就在本仓库 docs/media/deepseek/ 下;来源与许可见 docs/media/deepseek/SOURCE.md。

How to read this page / 怎么读这页 English first, then the same content in Chinese — not a summary, not a separate half. Tables keep both languages in one cell. 每段英文在前、同一内容的中文紧跟在后;不是摘要,也没有单独的"中文版"。表格把两种语言放进同一格。


At a glance · 一眼看懂

Problem
问题
A long agent run looks the same whether it is working, waiting for input, or already errored — and the token bill lives in yet another window.
长时间运行的 Agent,在工作、等输入、已经出错这三种状态下看起来一样;而 token 账单又在另一个窗口里。
Fix
做法
A desktop pet + an MCP server over one shared data directory: session state drives the animation, a local ledger tracks usage, reminders fire on a schedule.
桌宠加一个 MCP 服务,共用同一份数据目录:会话状态驱动动作,本机台账记录用量,提醒按时触发。
Cost
代价
Python 3.10+ and PySide6 for the desktop; the agent path only needs requirements-agent.txt. Dining features need McDonald's / Luckin tokens.
桌面要 Python 3.10+ 与 PySide6;只走 Agent 路径只需 requirements-agent.txt。餐饮功能需要麦当劳/瑞幸凭据。
Not
不是
Not a hosted service, not a model provider, not an ordering bot that pays for you.
不是托管服务,不是模型提供方,也不是替你付款的下单机器人。

What it does · 能做什么

Feature · 功能 Desktop entry · 桌面入口 Say this to the agent · 对 Agent 说
Companion & skins
陪伴与换装
Character & actions → default / McD outfit "DeepSeek 娘穿麦麦套装"
Session state
会话状态
Sessions & quota → session progress "用 McPet 看当前会话状态"
Usage & balance
用量与余额
Sessions & quota → local ledger, Orca cache, DeepSeek balance "余额还有多少"
McDonald's
麦当劳
McDonald's → coupons, store, live quote "今天有什么麦当劳优惠?先查一下"
Luckin coffee
瑞幸
Luckin → store, item, spec, preview price "选一杯拿铁,先预览价格"
Nutrition ledger
营养记账
Nutrition → pick item, confirm servings "查询巨无霸营养,我吃了 1.5 份"
Reminders
定时提醒
Settings → time & location → lunch / coffee "饭点改成 12:15"

Every order, coupon claim and ledger write shows a concrete summary first and waits for explicit confirmation. The default order guardrails are 100 CNY per order and 3 orders per day, re-checked at confirmation time.

每一次领券、下单和记账都先展示具体摘要,等明确确认后才执行。默认订单护栏是单笔 100 元、每日 3 单,确认时再复核金额。

The problem · 问题在哪

An agent session is a black box with a progress bar:

agent  ──▶ working?          (no output for 6 minutes)
       ──▶ waiting for input? (a prompt is on screen)
       ──▶ stopped?           (idle / Stop)

idle and Stop are session states, not success. The pet shows the state it can actually read from logs or host events, and keeps unreadable fields unknown instead of guessing.

Agent 会话是个带进度条的盲盒:

agent  ──▶ 在工作?        (六分钟没有输出)
       ──▶ 在等输入?      (屏幕上有提示)
       ──▶ 已停止?        (idle / Stop)

idle 与 Stop 是会话状态,不是成功。桌宠只显示它真能从日志或宿主事件读到的东西,读不到的字段保持未知。

The idea · 思路

From the project's README.

Comments (0)

Sign in to join the conversation.

No comments yet.