Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
作者提供的说明(英文原文)
Local-first dev utilities: hash, base64, JWT, UUID, cron, URL codec. No network or telemetry.
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。
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。
选择你使用的 AI 应用,然后复制设置。
你需要
你用的是哪个应用?
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": {
"toolars": {
"command": "npx",
"args": [
"-y",
"@toolars/cli",
"-y",
"-p",
"@toolars/cli",
"toolars-mcp"
]
}
}
}在终端中运行这条命令
claude mcp add toolars -- npx -y @toolars/cli -y -p @toolars/cli toolars-mcp粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"toolars": {
"command": "npx",
"args": [
"-y",
"@toolars/cli",
"-y",
"-p",
"@toolars/cli",
"toolars-mcp"
]
}
}
}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。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"toolars": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@toolars/cli",
"-y",
"-p",
"@toolars/cli",
"toolars-mcp"
]
}
}
}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。
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。
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6日
注册名称已通过 GitHub 账户 aixtral 验证,该账户拥有此仓库
我们还没有从 GitHub 读取这个仓库
我们还没有从 GitHub 读取这个仓库
它在你的电脑上运行,没有互联网端点
在你的电脑上运行,改用本地设置
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。
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。
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)。
GitHub 仓库最后更新于–,最新版本(0.1.1)于41天前进入注册中心。
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。
同一分类中有:Spec Driven Development, Remote Desktop Commander, Zoo MCP
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
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
An MCP server that provides access to the Zoo API for various CAD operations and tools.
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。
Use Toolars online for PDF, image, text, and developer tools directly in your browser.
This directory is the source of truth for the code Toolars publishes publicly. It is assembled into a distributable tree by scripts/export-open-source.mjs in the private repository; nothing here is built or bundled into the website.
Every push and every day at 03:17 UTC, the Privacy proof workflow rebuilds both packages from this tree alone, re-runs their test vectors, and re-captures the privacy-proof receipt against the live site. The badge is green only when the latest receipt's outcome is verified.
A failed run is not automatically a privacy finding. The harness distinguishes two outcomes: violation means a measured request crossed the local-processing boundary (the workflow fails on the first attempt, without retrying); inconclusive means the run could not be completed — usually a network problem between the runner and the site — and says nothing about the site in either direction. Inconclusive runs are retried; a persistent one fails the job with an explicit notice. Read the run log before reading anything into a red badge.
| Path | What it is |
|---|---|
evidence-harness/ | A runnable harness that drives the public website in headless Chromium and records every network request, including dedicated Web Worker fetches. It is the tool behind the claim that files never leave the browser. |
packages/cli/ | @toolars/cli — a local stdio CLI and MCP server (hash, base64, JWT decode, UUID/ULID, timestamp, cron, URL encode/decode), shipped as a prebuilt bundle in dist/. |
packages/local-tools/ | @toolars/local-tools — a smaller local stdio CLI and MCP server (JSON formatting, SHA-256), shipped as runnable source. |
All of it is MIT-licensed (see LICENSE). THIRD-PARTY.md states the exact scope and lists third-party components.
This is an open-core distribution, not the whole product. Not included:
packages/*/vendor/ carries the site runner cores these two packages import (15 files for the CLI, 2 for local-tools), because without them the packages could not be rebuilt or re-tested here. The packages/cli bundle also inlines compiled copies of the same cores, since it is the artifact prepared for a future @toolars/cli npm release. Both are covered by this repository's MIT license;That boundary is the point: the harness verifies the deployed site, so independent verification does not require the site's source. See evidence-harness/README.md for exactly what the harness does and does not prove.
cd evidence-harness
npm install npx playwright install chromium npm run verify -- --base-url https://toolars.com
The run writes privacy-proof-receipt.json and exits non-zero if an unexpected request leaves the site's origin, if any request carries the canary marker, or if any request uploads data. Disclosed shell requests are counted separately. A missing or changing release identity also prevents a verified result.
Both packages are published on npm under the @toolars scope: npm install -g @toolars/cli gives you toolars + toolars-mcp, and npm install -g @toolars/local-tools gives you toolars-local + toolars-local-mcp. The bins do not collide.
cd packages/cli
node dist/cli.mjs hash --sha256 "hello" node dist/mcp.mjs # stdio MCP server
dist/ is committed and copied from the private build output, so running it needs no install and no build. Review that bundle if you want to see the code you would actually execute: it is unminified, and its section comments name the source file each part came from.
You can also rebuild it here and re-run the same test vectors the private workspace runs. Each package carries the exact site core closure it depends on under vendor/, so neither needs the private repository:
cd packages/cli
npm install # esbuild + vitest, declared as devDependencies npm run build # writes dist/ from src/ plus vendor/ npm test # shared digest vectors and CLI/MCP boundary tests
cd packages/local-tools
npm install # esbuild + prettier node build.mjs # regenerates src/json-core.mjs from vendor/ node --test test/*.test.mjs
vendor/ holds copies of the site runner cores these packages import, produced by the exporter and byte-identical to the private ones. They are published as source on purpose: a reader who can rebuild the package and re-run its vectors can check the results instead of trusting them. Rebuilding in the exported layout changes source-path comments in the CLI bundles and local-tools generated core. Compare executable content separately from these comments; a rebuild is not claimed to reproduce the original manifest byte for byte.
Both packages are local-stdio only. Their inputs and results enter the calling client's context and may reach that client's model provider; the website's privacy promise does not extend to the client. Each package's README states this scope.
Open an issue on the public repository. Security reports follow the process in that repository's SECURITY.md.
MIT — see LICENSE. Scope and third-party components: THIRD-PARTY.md and evidence-harness/fixtures/PROVENANCE.md.