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)
What changed in the packages, APIs and services your code depends on, and the fixes it needs.
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://api.since.dev/mcp在终端中运行这条命令
claude mcp add --transport http since-dev https://api.since.dev/mcp粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"since-dev": {
"url": "https://api.since.dev/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": {
"since-dev": {
"type": "http",
"url": "https://api.since.dev/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。
ChatGPT 只能连接互联网上的服务器:
https://api.since.dev/mcp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。
服务器要求先登录,才会列出它的工具。
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 since-dev 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
在 286 毫秒内响应(HTTP 401)
声明支持 OAuth 登录,你的 AI 应用可以带你去登录
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 3 项适用的基础检查中通过了 3 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(4.1.0)于今天进入注册中心。
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, Recta
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.
Cheaper cross-chain stablecoin routes for AI agents; 5 % of the verified saving, nothing if none.
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。
What changed in the packages, APIs and services your code depends on, and the fixes it needs.
Product · Documentation · MCP reference
Since.dev tracks every package in 18 registries, every AI model's retirement on each platform that serves it, API changes and runtime end-of-life dates, each with its source. This repository connects your agent to it: a hosted remote MCP server, plus a plugin for Claude Code, Cursor and Codex that tells the agent when to use it. There is no server code to install or run.
Every install signs in through your browser (OAuth). You approve the client for one Since.dev workspace, and no key is stored in a file. A free account works.
Install the plugin, which adds the server and four skills:
/plugin marketplace add since-dev/mcp
/plugin install since-dev@since-dev
Or add only the server, then run /mcp to sign in:
claude mcp add --transport http since https://api.since.dev/mcp
Click Add to Cursor above, or add this repository as a plugin to get the server, the skills and a rule that checks dependency, model and runtime changes.
Click Install in VS Code above, or run:
code --add-mcp '{"name":"since","type":"http","url":"https://api.since.dev/mcp"}'
Install the plugin:
codex plugin marketplace add since-dev/mcp
Or add only the server and sign in:
codex mcp add since --url https://api.since.dev/mcp
codex mcp login since
gemini extensions install https://github.com/since-dev/mcp
The extension asks for a workspace API key (see below). To sign in through the browser instead, add the server directly:
gemini mcp add --transport http since https://api.since.dev/mcp
Add a remote server with this URL over Streamable HTTP and sign in when the client asks:
https://api.since.dev/mcp
For a client that reads an mcpServers file, see examples/mcp.json. For VS Code's .vscode/mcp.json, see examples/vscode-mcp.json.
Some clients can't sign in through a browser. For those, create a key and send it as a Bearer token:
sc_live_ and is shown once.Authorization: Bearer in your client's secret settings. Never commit a key.A key reaches only its own workspace, over the API and MCP, and any member with a seat can revoke it under API keys.
whats_changed and check_code only read, so your client runs them without asking. fix_code opens fix pull requests, so your client asks before every call.since-before-upgrading, since-model-retirements, since-what-changed, since-ci-and-runtimes. They tell the agent when to check: before upgrades, when code names a model, before writing against something it learned in training, and when editing CI or runtimes.tools/list returns each tool's full input schema.
whats_changed and check_code only read, and both work on every plan; checking a connected repository (check_code with repo) is on the Business and Platform plans. fix_code writes: with code it returns a patch, on the Team plan and up, and with repo it opens fix pull requests for a person to review, on the Business and Platform plans. Each call costs credits; a bad input or a call the plan does not include is refused with a plain sentence and costs nothing.
| Tool | What it does | Key inputs |
|---|---|---|
whats_changed | What changed in a package, a provider, a model, a runtime or a connected repository, since your training cutoff, a date or a version, with severity, dates, replacements and cited evidence. Also lists a provider's current models to use in new code (current=true), gives one change in full with its migration steps (id), answers whether something is still supported, and lists upcoming retirements (as iCalendar too). | One of package (any spelling or purl), provider, model, repo, surface or id (one change, as a check_code finding names it); optional since (a date, a version or cutoff: ), current (with provider), within (for example 90d), kind, severity, reaches (with repo), cursor, limit |
check_code | Checks code against what changed, before you rely on it: a snippet you just wrote, a model id, a package version, a manifest or a whole repository. Returns what is affected, where (file and line), how severe, how many days are left, the replacement and the source. Anything Since cannot settle is needs_review, never a guess. It only reads. | One of model, package (with a version), surface, code (with language), manifest or lockfile (path and content), repo, or repoUrl (a public GitHub repository); optional state, severity, detail (summary or full), limit, cursor |
fix_code | Fixes what check_code found: the patch for a snippet, in seconds, or a connected repository's fix pull requests, one per provider, each checked before it opens. Nothing merges: every pull request waits for a person to review it. Fixes count against the month's fix pull requests, not credits. | One of repo or code (with language or path); optional state, severity, detail, limit |
Each step shows the name and arguments the agent sends in a tools/call request.
{ "name": "check_code", "arguments": { "model": "<model id>" } }
The result lists the model's status (supported, deprecated, retired or unknown) and a finding for each change that applies, with its severity, daysLeft, replacement and cited source.
{ "name": "whats_changed", "arguments": { "package": "npm:react", "since": "18.2.0" } }
Each change carries its canonical surface, kind, severity, dates, replacement and evidence. package.releasesBehind says how far behind the version is, and coverage says which sources were read and how fresh they are. Page with nextCursor.
{ "name": "whats_changed", "arguments": { "provider": "anthropic", "within": "90d" } }
{ "name": "fix_code", "arguments": { "repo": "<repository id>" } }
Treat unknown and needs_review as not settled, never as fine.
The files in this repository are MIT licensed. See LICENSE.