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)
Read-only VNISH firmware metadata, Ninja guidance and input-based mining economics calculations.
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 …”。
适用于网页版、桌面版和手机版 Claude:
https://vnish.global/ai/mcp/index.php在终端中运行这条命令
claude mcp add --transport http vnish-global-operator https://vnish.global/ai/mcp/index.php粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"vnish-global-operator": {
"url": "https://vnish.global/ai/mcp/index.php"
}
}
}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": {
"vnish-global-operator": {
"type": "http",
"url": "https://vnish.global/ai/mcp/index.php"
}
}
}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。
ChatGPT 只能连接互联网上的服务器:
https://vnish.global/ai/mcp/index.php菜单名称可能因套餐和版本不同而略有差异。
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。
我们还没有读取这个服务器的工具列表;我们会轮流检查各个端点。
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
注册名称已通过 GitHub 账户 vnish-global 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
尚未检查;我们会轮流检查各个端点
尚未检查;我们会轮流检查各个端点
尚未检查;我们会在检查端点时读取工具列表
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。
它在作者的云端运行。通过互联网连接,无需安装,但你发送的内容会经过作者的服务器。
GitHub 仓库最后更新于–,最新版本(1.0.0)于43天前进入注册中心。
支持 MCP 的应用,例如 Claude Desktop、Claude Code、Cursor 和 VS Code。它通过 URL 连接,所以接受远程 MCP 服务器的应用也可以使用,例如开启开发者模式的 ChatGPT。
同一分类中有:Spec Driven Development, Remote Desktop Commander, Python 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.
Local-only Python MCP server for the Files.com API.
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。
The remote MCP endpoint and connection guides are now published:
https://vnish.global/ai/mcp/index.phpPublic HTTPS verification with the official MCP JavaScript SDK 1.30.0 passed all 7 checks: PHP execution, protocol initialization, tool listing, resource access and one call to each of the three read-only tools. The negotiated protocol was 2025-11-25. The two guide files, their stylesheet and the catalog snapshot matched the prepared release by SHA-256 and byte count.
Connection inside native ChatGPT and Claude accounts has not been tested. This result does not claim app-directory approval or physical miner testing.
The original pre-deployment snapshot is preserved unchanged below. Its planned-publication wording records the earlier state.
Status: NOT YET DEPLOYED. This is a reviewed source package and deployment candidate. The HTTPS endpoint and new connection-guide URLs below are planned. This README does not claim that ChatGPT or Claude has connected to the service or approved a directory listing.
The PHP server exposes three read-only tools using public VNISH Global catalog metadata, selected VNISH Ninja references and an ROI ASIC arithmetic scenario. It does not connect to miners, install firmware, run commands, fetch arbitrary URLs, use credentials or store tool arguments. Existing Operator skills and earlier MCP implementations are preserved.
The six files under public/ are an additive deployment for vnish.global:
| File | Purpose |
|---|---|
ai/mcp/index.php | Streamable HTTP endpoint |
ai/mcp/catalog-snapshot.json | Dated public catalog snapshot |
ai/mcp/.htaccess | DirectoryIndex index.php in this new directory only |
ai/connect/index.html | English connection guide |
ai/connect/connect.css | Isolated guide styles |
ru/ai/connect/index.html | Russian connection guide |
Planned endpoint: https://vnish.global/ai/mcp/index.php.
Planned guides: https://vnish.global/ai/connect/ and https://vnish.global/ru/ai/connect/.
Deploy only public/ paths, after checking for existing files. Verify PHP execution and POST behavior on the hosting runtime. No existing site root, shared navigation, styles, JavaScript, sitemap, robots or security configuration is part of this package. GET and DELETE on the MCP endpoint return 405; connection guides are separate HTML pages. Test and data-source files do not belong in the web root.
lookup_firmware: exact catalog model ID, with optional board code and version. Returns filename, size, SHA-256 and Global source URLs. Multiple matching records stay ambiguous. A catalog record is not evidence of compatibility, authenticity or successful hardware testing.find_ninja_reference: fixed EN/RU guides for control-board identification, installation SSH errors and return-to-stock preparation. It does not determine whether a recovery method applies to a particular device.calculate_roi_scenario: gross pre-fee hashrate, average wall power, modeled fee, gross daily hashprice, electricity tariff and operating hours. All inputs are supplied by the caller. Both revenue and energy scale by the same operating-hours assumption. The output is contribution after the modeled fee and electricity, excluding other expenses; it is not net profit, ROI or a forecast.The server also exposes three resources: the model ID index, snapshot provenance and scenario methodology. Resource URIs are identifiers used with MCP resources/read, not promised web pages.
data/source-catalog.json is a copy of the public catalog at https://vnish.global/api/v1/firmware-catalog.json. data/catalog-provenance.json records its retrieval timestamp and SHA-256. The bundled snapshot contains 47 models and 148 build records. A snapshot is not a live availability check. No scheduled update process is included.
Selected identifiers and scenario numbers are sent to the server. Application code does not persist arguments; standard hosting/CDN logs may record request metadata. Passwords, account data, miner addresses and configuration files are unnecessary.
Protocol versions: 2026-07-28, 2025-11-25, 2025-06-18, 2025-03-26. Modern requests use server/discover, per-request metadata and mirrored headers. Legacy clients use initialize. The server creates no session ID and returns JSON instead of SSE. It provides no subscriptions, elicitation, sampling, batch requests or device operations.
Present Origin headers must exactly match the three project domains, https://chatgpt.com or https://claude.ai; absent Origin is permitted for server clients. Inputs are bounded to 32 KiB and depth 32. IDs must be strings or numbers whose integer value is in the safe range from -(2^53-1) to 2^53-1, so request correlation cannot be silently rounded.
Requirements: Node.js 20.10 or newer; Python 3.9 or newer for builders. Runtime PHP hosting is separate from the test dependencies.
Pinned test dependencies:
@php-wasm/universal 3.1.54@php-wasm/node-8-4 3.1.54@modelcontextprotocol/sdk 1.30.0A pnpm lockfile pins transitive dependencies. From this directory:
pnpm install --frozen-lockfile
pnpm test
The tests execute the actual endpoint in PHP 8.4 WebAssembly, then start a temporary HTTP bridge bound to 127.0.0.1 and call it with the official SDK's StreamableHTTPClientTransport. The bridge closes when testing finishes. Tests cover tools, resources, current protocol metadata, origin validation, bounded inputs, scenario arithmetic, negative results and malformed/fractional/oversized request IDs. They need no production account, credentials or miner.
Generated receipts go in ignored test-results/. Tests do not access arbitrary network destinations; example invalid URLs are negative-test fixtures. No browser scripts or private QA artifacts are included.
Rebuild the public snapshot and guides from the bundled public source:
pnpm run build
A successful local test does not establish public hosting or an end-user client connection. After deployment, verify the public HTTPS endpoint and then test actual client connections.
The connection pages' canonical/language URLs, their local connect.css, and the displayed MCP URL will work only after the public/ files are published. Guide links to the current public catalog, existing Ninja and ROI ASIC references, official setup documentation and the existing Operator repository are independent of that deployment. The repository link deliberately points to the existing repository root until this package is published.
Setup guidance was checked against official documentation on 2026-09-17. Account and workspace policy can affect available connection options.