Pls Touch Grass

Name: Pls Touch Grass
Author: Rinava

作者：Rinava版本 0.3.0i★ –i在你的电脑上运行

注册名称io.github.Rinava/pls-touch-grass

安装

概览

作者提供的说明（英文原文）

An MCP server that knows how long it's been since you touched grass

PopNx MCP 的三大支柱

这三项永远不会合并成一个分数：意见、测量和人气是不同的东西。

用户评价i暂无评价评价功能将在后续阶段与账户功能一起开放
我们的基础检查i1 项中通过 1 项依据公开信息检查；尚未扫描代码
使用情况i尚未统计★ –

它能帮你做什么

我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。

连接方式: 在你的电脑上运行
传输方式: stdio
分类: 其他

安装

选择你使用的 AI 应用，然后复制设置。

安装之前

基础检查：1 项中通过 1 项
在你的电脑上运行，拥有你的用户账户的权限（文件、网络、应用）。
需要你自己填写： GRASS_STATE_FILE

你需要

npm已安装 Node.js（用于 npx 命令）

你用的是哪个应用？

ClaudeClaude CodeCursorVS CodeChatGPT

在上方选择你的应用。我们会记住它，下次按钮会显示“安装到 …”。

打开 Claude Desktop → Settings → Developer → Edit Config，把下面的内容粘贴到 claude_desktop_config.json

{
  "mcpServers": {
    "pls-touch-grass": {
      "command": "npx",
      "args": [
        "-y",
        "pls-touch-grass-mcp"
      ],
      "env": {
        "GRASS_STATE_FILE": "your-value"
      }
    }
  }
}

在终端中运行这条命令

claude mcp add pls-touch-grass --env GRASS_STATE_FILE=your-value -- npx -y pls-touch-grass-mcp

运行前请先填写空着的值。

在 Cursor 中打开

或者自己复制设置

粘贴到项目中的 .cursor/mcp.json，或 ~/.cursor/mcp.json

{
  "mcpServers": {
    "pls-touch-grass": {
      "command": "npx",
      "args": [
        "-y",
        "pls-touch-grass-mcp"
      ],
      "env": {
        "GRASS_STATE_FILE": "your-value"
      }
    }
  }
}

浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。

在 VS Code 中打开

或者自己复制设置

粘贴到项目中的 .vscode/mcp.json

{
  "servers": {
    "pls-touch-grass": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "pls-touch-grass-mcp"
      ],
      "env": {
        "GRASS_STATE_FILE": "your-value"
      }
    }
  }
}

浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。

这个服务器在你的电脑上运行，而 ChatGPT 只能连接互联网上的服务器。请选择其他应用。

你的选择通过 Cookie 保存在这台设备上，无需账户。

需要你填写的值

GRASS_STATE_FILE env
Path to the JSON file holding your grass-touching history. Defaults to ~/.pls-touch-grass.json

这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同，请以 README 为准。

它能做什么

它在你的电脑上运行；安装后，工具列表会显示在你的 AI 应用中。

基础检查i

我们检查的是任何人都能看到的信息；我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日

✓ 通过
作者与仓库一致
注册名称已通过 GitHub 账户 Rinava 验证，该账户拥有此仓库
… 尚未检查
仍在维护
我们还没有从 GitHub 读取这个仓库
… 尚未检查
有许可证
我们还没有从 GitHub 读取这个仓库
– 不适用
端点有响应
它在你的电脑上运行，没有互联网端点
– 不适用
使用 OAuth 登录
在你的电脑上运行，改用本地设置
… 尚未检查
工具描述中没有隐藏指令
它在你的电脑上运行；我们只能读取互联网端点的工具列表

与“其他”中的其他服务器比较i

星标数多于此分类中其他 18354 个服务器的 0%
发布到注册中心的时间比此分类中其他 18354 个服务器的 21% 更新
基础检查结果等于或优于此分类中其他 18354 个服务器的 100%

与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时，星标数才计入。

常见问题

Pls Touch Grass 安全吗？

它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前，请先看看这个服务器要求哪些权限，并使用只具备所需权限的密钥。

它需要哪些密钥或代码？

注册中心没有列出必需的密钥，不过有些服务器会在首次使用时要求你登录。

它在我的电脑上运行还是在云端？

它在你的电脑上运行。你的 AI 应用会在需要时启动它，数据不会经过作者的服务器（除非它自己调用在线服务）。

它还在维护吗？

GitHub 仓库最后更新于–，最新版本（0.3.0）于3个月前进入注册中心。

哪些应用可以使用它？

支持 MCP 的应用，例如 Claude Desktop、Claude Code、Cursor 和 VS Code。你的电脑上还需要有 Node.js。

有什么可以替代它？

同一分类中有：Spec Driven Development, Remote Desktop Commander, Qemu MCP

替代选择

同一分类中其他作者的服务器，按星标数排序。

如何看卡片：★ GitHub 星标i三大支柱：评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i

Spec Driven Development

作者：formulahendry

MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.

在你的电脑上运行★ 440MIT14个月前更新

安装

Remote Desktop Commander

作者：desktop-commander

Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.

在云端运行★ 211前天更新

安装

Qemu MCP

作者：0xmortuex

Boot and drive QEMU VMs from an AI agent: keys, screenshots, serial. No guest agent.

在你的电脑上运行今天更新

安装

作者的 README

打开 README

按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。

pls-touch-grass-mcp

An MCP server that knows how long it's been since you touched grass, and won't shut up about it.

You log grass, it tracks indoor time, it catches you raging at your AI and sends you to the nearest park, weather checked first. A joke, but one that teaches MCP: five tools, shared state, and descriptions that make the model call them unprompted.

Install

Claude Code

claude mcp add pls-touch-grass -s user -- npx -y pls-touch-grass-mcp

-s user, or the server only exists in the directory you ran that from. Working from a clone, skip the command: there's a .mcp.json in the root and Claude Code offers it when you open the folder.

Cursor

Or the JSON, in ~/.cursor/mcp.json (global) or .cursor/mcp.json (per project):

"mcpServers": { "pls-touch-grass": { "command": "npx", "args": ["-y", "pls-touch-grass-mcp"] } } }

Claude Desktop

.mcpb from Releases, then Settings → Extensions → Install Extension. Dependencies are bundled; installs offline. By hand it's the same JSON as Cursor, in claude_desktop_config.json (~/Library/Application Support/Claude/ on macOS, %APPDATA%\Claude\ on Windows).

claude.ai

No. Custom connectors want a remote server over HTTPS; this is stdio on your own machine. HTTP transport is the first item under Homework.

From source

git clone https://github.com/Rinava/pls-touch-grass-mcp

cd pls-touch-grass-mcp npm ci && npm run build

Point any config above at node /absolute/path/to/dist/index.js instead.

npx -y downloads the package on first start — live audiences and venue wifi deserve a warm cache. Published on npm as pls-touch-grass-mcp, in the MCP Registry as io.github.Rinava/pls-touch-grass.

The five tools

Tool	What it does	When the model calls it
`touched_grass`	Logs that you went outside	"back from the park"
`gotta_go`	Do you need to go out?	"can I keep going?"
`frustration_detector`	Catches the rage and escalates	on its own, when you snap; that's the point
`grass_conditions`	Real weather (Open-Meteo, no key)	before sending you outside
`where_to_touch_grass`	Nearby parks with distances (OpenStreetMap)	"where should I go?", or when you name a place

Nobody teaches the model to detect frustration: the tool description does it alone. Type "NOTHING WORKS!!!" and watch.

Zero configuration

Location resolves fresh from your IP (ipwho.is) each run — city-level on purpose. Detection down: last known spot, then the Obelisco, like everyone else. Name a neighborhood or a street address to where_to_touch_grass and it sticks: known neighborhoods resolve offline, anything else is geocoded once through OpenStreetMap's Nominatim, and only the words you typed leave your machine. A place you named outranks a guess from your IP, until you name a different one.

Indoor tolerance defaults to a merciless 2 minutes; ask the model for clemency and it passes a bigger threshold_minutes to gotta_go. State is one JSON file, ~/.pls-touch-grass.json. Touching grass resets the rage counter. That's how absolution works.

The demo curse

Live demos summon it: the venue wifi dies the moment you say "watch this". Add --demo to the args and every answer goes canned — 22°C and clear, the Obelisco, the curated spots. Zero network calls, zero surprises. Off by default; real life should stay real.

Follow the workshop

The repo is tagged by step:

git checkout step-1   # stdio server + touched_grass

git checkout step-2 # + gotta_go (state and threshold) git checkout step-3 # + frustration_detector (the demo) git checkout step-4 # + grass_conditions (network with fallback) git checkout step-5 # + where_to_touch_grass and this README npm ci && npm test

Every step compiles and passes its own tests. Two dependencies total: the official MCP SDK and zod.

Homework

HTTP transport: createMcpHandler in SDK v2 serves the same factory over fetch, and it's what gets you onto claude.ai
Per-project thresholds instead of a global one
Streak tracking for consecutive grass days
Demo spots for your own city, via PR: one object per park in src/lib/demo.ts, with an opinion

Release

npm run build:mcpb packs dist and the production dependencies into pls-touch-grass-mcp- .mcpb, the file that goes on a GitHub Release. The manifest is generated from package.json; nothing to keep in sync by hand.

Contributing

Issues and PRs welcome — the Homework list is seeded as open issues, several tagged good first issue. See CONTRIBUTING.md for setup and the house rules (there are few, but the demo-spot opinion is mandatory).

License

MIT. The grass is free and so is this.