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)
For advisors: every federal plan on your clients' loan files, RAP to PSLF. Sign in or use an API key
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。
X-Api-Key标为机密的值是你账户的钥匙。千万不要分享给别人,也不要粘贴到聊天里。
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 …”。
这个地址需要在请求的 header 中放入密钥,而添加连接器的菜单无法设置。请改用 Claude Code、Cursor 或 VS Code。
在终端中运行这条命令
claude mcp add --transport http verified-engine https://mcp.finology.tech/mcp --header 'X-Api-Key: your-value'运行前请先填写空着的值。
粘贴到项目中的 .cursor/mcp.json,或 ~/.cursor/mcp.json
{
"mcpServers": {
"verified-engine": {
"url": "https://mcp.finology.tech/mcp",
"headers": {
"X-Api-Key": "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。
粘贴到项目中的 .vscode/mcp.json
{
"servers": {
"verified-engine": {
"type": "http",
"url": "https://mcp.finology.tech/mcp",
"headers": {
"X-Api-Key": "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。
这个地址需要在请求的 header 中放入密钥,而添加连接器的菜单无法设置。请改用 Claude Code、Cursor 或 VS Code。
你的选择通过 Cookie 保存在这台设备上,无需账户。
X-Api-Key 机密 HTTP header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7日
注册名称已通过域名 finology.tech 验证(注册中心会核实域名所有权)
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 119 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。
你需要填写:X-Api-Key
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(0.2.5)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, AvaHR
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.
Read hiring data and create inactive assessment and questionnaire drafts for human review.
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。
Language models guess at federal student loan rules. This doesn't.
@finology/mcp-server connects Claude, GitHub Copilot, Cursor, or any MCP client directly to Finology Software's parity-verified federal repayment engine. Every number is computed server-side against current federal rules and returned with a parityVerified flag. No formulas are approximated, and none are hardcoded into this package.
Federal repayment eligibility is not arithmetic, it is regulation, and it is the part models get wrong most confidently:
compare_federal_student_loan_repayment_plansCompares a balance across every plan the borrower is actually eligible for — Standard, tiered Standard, RAP, IBR, PAYE and ICR — returning monthly payment, lifetime cost, projected forgiveness and tax on forgiveness for each, plus the plans that are excluded and the eligibility rule that excludes them.
compare_married_filing_jointly_vs_separately_student_loansFor a married borrower, prices the filing-status decision. Filing separately removes the spouse's income from the income-driven payment calculation, which routinely moves the lifetime loan cost by tens of thousands of dollars.
When both spouses carry federal loans, pass the spouse's own federal balance as well: under joint filing the household payment is split by each spouse's share of the combined balance (34 CFR 685.209(g)(1)(i)), and leaving it out overstates the borrower's payment.
It answers half the question on purpose, and says so. The engine models the loan side exactly and does not model the tax cost of filing separately — lost credits, worse brackets, community-property splits — which is often large enough to reverse the answer. Returning only the favorable half as if it were the whole decision would be the same confident half-truth this server exists to replace.
estimate_rap_monthly_paymentThe RAP monthly payment alone, for the common case. Requires the borrower's AGI and refuses to proceed without it — an assumed zero returns the $10/month RAP floor, which reads exactly like a real answer.
Every answer ends with the next rung, and finology_service_info returns it structured:
POST https://api.finology.tech/v1/keys/sandbox, 25 calls/month), then the keyed endpoints with X-Api-Key. The keyed rung is the answer-of-record surface: every answer is persisted with its rule version and an inputs hash before it is served. Production access is self-serve: Developer, US $49 a month for 2,000 calls, or Platform, US $499 a month for 100,000 calls. Docs: https://finology.tech/developers/https://mcp.finology.tech/mcp. Sign in with a Finology Software login (OAuth), which is how Claude, ChatGPT and Microsoft Copilot connect, or send the X-Api-Key header. Every answer is logged and carries a verify link.Claude (web, desktop and mobile) — one click. Finology Software is listed in the Claude Connectors Directory. Add the connector there and sign in with a free Finology Software login (no card): 5 questions in total, no daily limit. Subscribers are unlimited, and a subscriber's login also lists their own clients in the same connector (list_my_clients, run_projection_for_client, get_balance_trajectory_for_client).
ChatGPT. Finology Software is in the ChatGPT plugin directory, with the same free login.
Microsoft Copilot. GitHub Copilot (including its free plan) adds the remote server once: copilot mcp add --transport http finology https://mcp.finology.tech/free, then signs in with a Finology Software login the first time it calls a tool; GitHub Copilot in Visual Studio Code uses the same address and sign-in. In Copilot Studio, add a Model Context Protocol tool with that URL and OAuth 2.0 Dynamic discovery; running the agent uses your firm's Copilot credits. Setup: https://finology.tech/finology-software-in-microsoft-copilot/
Everything below is the manual route, for other MCP clients or for a local install.
Claude Desktop — add to claude_desktop_config.json:
{
"mcpServers": { "finology": { "command": "npx", "args": ["-y", "@finology/mcp-server"] } } }
The local package needs no API key and no account: it calls the public calculation endpoints.
Not here, deliberately: the RAP/OBBB formulas, income-driven payment math, forgiveness and tax-bomb projection, plan-eligibility rules, and NSLDS parsing. Those run on Finology's servers.
That is not only about protecting the engine. Federal repayment rules move every year — poverty guidelines, tax brackets, RAP/OBBB implementation guidance, court challenges. A formula copied into an npm package is wrong within months and cannot be corrected in the installs that already exist. Calling a maintained endpoint is the only shape of this tool that stays true.
Here: the input schemas, the HTTP call, and error handling that passes the API's own field-level messages through to the model instead of flattening them into "request failed."
These tools return figures. They do not produce client-facing deliverables, save borrower records, parse NSLDS files, or track PSLF qualifying payments over time. For those, and for a book of borrowers kept current as the rules change: [finology.tech](https://finology.tech/?utm_source=mcp&utm_medium=ai_agent)
The local package is rate-limited per IP. Not financial advice.
MIT