原型 · 预览版:收录官方注册中心里的全部服务器,每 30 分钟同步一次。使用 Google 登录即可撰写评价。原型 · 每 30 分钟更新
PopNx MCP为使用 AI 的人准备的 MCP 服务器 · 用大白话讲清楚 登录

Cheap First

作者:robd-hub版本 0.1.0i★ 0i在云端运行官方网站 ↗
注册名称io.github.robd-hub/cheap-first

安装方法

能做什么

Answers cheaply first, then escalates to a premium model only when a checker fails.

作者提供的原文

  • 连接方式 在云端运行
  • 账号或密钥 不填密钥也能用 · 有 Authorization, X-Cheap-First-Paid 可以填上(通常能提高额度)
  • 分类 其他

为什么排在这里

列表按使用分(满分 100)排序,由以下部分组成

热度0/80GitHub 星标或每周下载量:较大的计 60,较小的计 20
新鲜度12/1230 天内有发布或代码更新得满分
仍在响应8/8上次检查时云端端点有响应
使用分20/100

质量: 基础检查通过 5/5 项(仅展示,不用于排序) · 上升势头: 暂无数据 (只用于“上升中”标签)

安装

选择你使用的 AI 应用,然后复制设置。

用一句话让 AI 帮你安装

请帮我安装 Cheap First。按照 https://mcp.popnx.com/i/robd-hub-cheap-first.md 操作,先把命令给我看,等我同意后再运行。

粘贴到能在你电脑上工作的 AI(如 Claude Code、Cursor)。它会读取我们的步骤,先给你看命令,等你同意。无需密钥,无需注册。 · 查看 AI 会读取的全文

通过 5/5 项检查 · 你的提问会经过开发者的服务器(cheap-first.vercel.app) · ✓ 命令已检查 2026年11月20日

第一步:你用哪个 AI?

选择你用的应用,步骤会显示在下方。

可连接 Claude · 约 2 分钟

不需要密钥,不用安装程序 · 在 claude.ai 或桌面应用设置一次,即可在聊天中使用 · 免费账户可添加 1 个 · Team/Enterprise 方案由所有者(Owner)添加

  1. 打开 Claude 的 Connectors 页面

    CustomizeConnectors

    CustomizeConnectors1Add2https://…3Add4
  2. 点击“Add” → “Custom” → “Web”,填写这 2 项

    把这里的内容复制到 Claude 中同名的输入框,然后点击 Continue

    Name · 名称Cheap First
    Remote MCP server URL · 地址https://cheap-first.vercel.app/mcp

    在 Claude 检测到的登录设置页点击 Continue · Authentication 按作者说明选择(Sign in now、Sign in when needed 或 No sign in),然后点击 Add

  3. 在聊天中开启

    +ConnectorsCheap First

  4. 试一试

    用 Cheap First 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Claude 说正在使用 Cheap First,或回答中出现 Cheap First 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同

能否连接 ChatGPT 取决于方案

完整 MCP:网页版 ChatGPT Business、Enterprise 或 Edu · Pro:以读取/获取权限连接,修改数据的工具可能用不了 · Plus:官方文档尚未确认,如果在插件页看到添加自定义 MCP 的选项可以试试 · 仅限网页

  1. 在网页版 ChatGPT 打开添加 MCP 的页面

    Plugins+Add custom MCP server

    ChatGPTPlugins1+2https://…3Create4
  2. 填写名称和地址

    输入名称,在 Connection 下粘贴地址,勾选风险提示,然后点击 Create as a plugin

    Name · 名称Cheap First
    Connection · 地址https://cheap-first.vercel.app/mcp
  3. 在聊天中选择 Cheap First

    开启新聊天,输入前先从插件列表中选择 Cheap First

  4. 试一试

    用 Cheap First 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 ChatGPT 说正在使用 Cheap First,或回答中出现 Cheap First 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 OpenAI 帮助页面核实 · 你的界面可能因版本略有不同

可连接 Claude Code · 约 2 分钟

不需要密钥,不用安装程序

  1. 打开终端

    Mac:按 Command+空格,输入 Terminal · Windows:打开 PowerShell

  2. 粘贴这条命令并按 Enter

    claude mcp add --transport http --scope user cheap-first https://cheap-first.vercel.app/mcp
  3. 在 Claude Code 中输入 /mcp

    应能看到 Cheap First 在列表中并显示已连接

  4. 试一试

    用 Cheap First 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Claude Code 说正在使用 Cheap First,或回答中出现 Cheap First 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Anthropic 帮助页面核实 · 你的界面可能因版本略有不同

可连接 Cursor · 约 2 分钟

不需要密钥,不用安装程序

  1. 点击这个按钮打开 Cursor

    在 Cursor 中打开

    Cursor 会请你确认安装,点击 Install

  2. 在 Cursor 中以 Agent 模式打开 AI 聊天

  3. 试一试

    用 Cheap First 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 Cursor 说正在使用 Cheap First,或回答中出现 Cheap First 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 Cursor 帮助页面核实 · 你的界面可能因版本略有不同

可连接 VS Code · 约 2 分钟

不需要密钥,不用安装程序

  1. 点击这个按钮打开 VS Code

    在 VS Code 中打开

    VS Code 会请你确认安装,点击 Install

  2. 在 VS Code 中以 Agent 模式打开 AI 聊天

  3. 试一试

    用 Cheap First 告诉我你能帮我做什么,然后做一件给我看

    ✓ 如果 VS Code 说正在使用 Cheap First,或回答中出现 Cheap First 的工具名称,就表示连接成功

菜单名称已于 2026年11月11日 对照 VS Code 帮助页面核实 · 你的界面可能因版本略有不同

其他支持 MCP 的应用(如 Windsurf、Zed、Gemini CLI)添加方式各不相同,请在应用列表中查看

查看所有支持 MCP 的应用 →

通过网页连接的应用使用的地址

地址https://cheap-first.vercel.app/mcp

JSON 设置在下方“开发者专用”中

卡住了?问问正在用 Cheap First 的人💬 讨论 →
开发者专用:终端命令和 JSON 设置

通过网页连接

Claude Code

claude mcp add --transport http --scope user cheap-first https://cheap-first.vercel.app/mcp

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "cheap-first": {
      "url": "https://cheap-first.vercel.app/mcp"
    }
  }
}

VS Code (mcp.json)

{
  "servers": {
    "cheap-first": {
      "type": "http",
      "url": "https://cheap-first.vercel.app/mcp"
    }
  }
}

需要你填写的值

  • Authorization 机密 HTTP header
    可选的预付费密钥,格式为 Bearer cf_...,用于付费额度。留空则使用免费演示。
    查看英文原文Optional prepaid key as Bearer cf_... for paid limits. Leave empty for the free demo.
  • X-Cheap-First-Paid HTTP header
    设为 1 可请求 x402 付款质询(Base 上 $0.005 USDC),而不是免费演示
    查看英文原文Set to 1 to request the x402 payment challenge ($0.005 USDC on Base) instead of the free demo.

说明为自动翻译,变量名保持原样;不确定时请查看原文。

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。

一览

每周下载量i
不在 npm/PyPI 上
连接方式
在云端运行
传输方式
streamable-http
最新版本i
0.1.0
分类i
其他
发布到注册中心i
2026年11月19日
代码最后更新i
2026年11月19日
最近检查
2026年11月20日
作者
robd-hub
注册状态
活跃

哪些应用可以使用它

依据这个服务器的连接方式和各应用自己的文档,检查于 2026年11月10日。

通过互联网连接,适用于

ClaudeChatGPTGooseLM StudioMsty StudioCherry StudioCursorVS Code + GitHub CopilotVisual StudioWindsurf (Devin Desktop)ZedKiroClineContinueJetBrains AI AssistantClaude CodeCodex CLIGemini CLIOpenCodenodecodeWarpLibreChatOpen WebUIRaycast

全部应用及其优缺点 →

它能做什么

工具名称和描述与服务器在 2026年11月20日 报告的完全一致。这些内容由作者撰写,不是我们写的;我们以纯文本形式显示。

  • ask_cheap_first

    Answer a question the Cheap First way: Dolphin answers, SERV Reasoning scores it, and a premium model answers instead only if the score misses the pass mark. Returns the answer, whether it escalated, the check score, the cost and the estimated saving. Free calls use the public de

  • cheap_first_stats

    Published benchmark figures. The main run is 143 questions: what Cheap First spent, what the premium model alone would have cost, how many Dolphin answered, and the average quality scores. Also includes the harder 97-question run. Does not call a model.

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20日

  • ✓ 通过

    作者与仓库一致

    注册名称已通过 GitHub 账户 robd-hub 验证,该账户拥有此仓库

  • ✓ 通过

    仍在维护

    代码最后更新于昨天

  • ✓ 通过

    有许可证

    使用 MIT

  • ✓ 通过

    端点有响应

    在 185 毫秒内响应(HTTP 200)

  • – 不适用

    使用 OAuth 登录

    无需登录

  • ✓ 通过

    工具描述中没有隐藏指令

    已读取 2 个工具描述,未发现针对 AI 的隐藏指令

我们检查什么 → · JSON

作者:申请重新检查或报告信息错误
给作者:在 README 中加徽章

如果你是作者,可把这段代码贴到 README。徽章会链接回本页。

[![Featured on PopNx MCP](https://mcp.popnx.com/badge/robd-hub-cheap-first.svg)](https://mcp.popnx.com/en/servers/robd-hub-cheap-first)

讨论

提问、分享技巧或报告无法使用。按“有帮助”数量排序。

还没有帖子,欢迎第一个提问或分享。

此处内容由用户撰写,不代表 PopNx 团队。

登录后发帖

用户评价

还没有评价

登录后写评价

与“其他”中的其他服务器比较i

  • 星标数多于此分类中其他 20958 个服务器的 0%
  • 发布到注册中心的时间比此分类中其他 20958 个服务器的 98% 更新
  • 基础检查结果等于或优于此分类中其他 20958 个服务器的 99%

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

Cheap First 安全吗?

它在 5 项适用的基础检查中通过了 5 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。

它还在维护吗?

GitHub 仓库最后更新于昨天,最新版本(0.1.0)于昨天进入注册中心。

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。它通过 URL 连接,因此也能用于接受远程 MCP 的应用,例如网页版 ChatGPT(Business 及以上为完整 MCP;Pro 为只读)。

有什么可以替代它?

同一分类中有:World Monitor, openstatus, Oh My Posh Validator, Strata, Jitsu

功能相近的其他选择

同一分类、其他作者、命令已通过检查。官方的排在前面,再按星标排序。

如何看卡片:↓ 每周下载量i★ 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

World Monitor

Real-time global intelligence on conflict, markets, energy, trade, shipping and cyber. 100 tools.

作者:koala73广泛使用★ 9.1万4天前更新

在云端运行✓ 服务方 · worldmonitor.app

openstatus

Manage monitors, status pages, incidents and maintenances in your openstatus workspace.

作者:openstatusHQ广泛使用★ 92762个月前更新

在云端运行✓ 服务方 · openstatus.dev

Strata

MCP server for progressive tool usage at any scale (see https://klavis.ai)

作者:Klavis-AI正在增长★ 582314个月前更新

在云端运行仓库已归档✓ klavis.ai

Jitsu

Manage Jitsu data pipelines: destinations, streams, connections, functions, live events.

作者:jitsucom广泛使用★ 51364个月前更新

在云端运行✓ 服务方 · jitsu.com
与相似项目对比
本项
Cheap First
World MonitoropenstatusOh My Posh ValidatorStrataJitsu
运行位置在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装在云端运行,无需安装
需要密钥否否否否否否
可在 Claude 中按 + 添加否否否否否否
基础检查5/54/44/44/41/44/4
许可证MITAGPL-3.0AGPL-3.0MITApache-2.0MIT
价格登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用登记中未列出费用

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。

Cheap First

Frontier quality at cheap-model prices.

Dolphin answers first. SERV Reasoning checks it. A premium model steps in only when the check fails.

Live results · Try it · Demo video · X post · How it works · Results · Run it yourself

Why

Most questions don't need a frontier model, but teams send everything to one because they can't tell in advance which answers will go wrong. Cheap First makes that call per question: the cheap model answers, a cheap checker grades it, and only failing answers are paid for at frontier prices.

How it works

flowchart LR

Q([Question]) --> D[" 1. Dolphin answers DeepSeek V4.1 Flash $0.07 / M output tokens"] D --> C{" 2. SERV checks GPT-6 Luna scores 0 to 100"} C -- "75 or more" --> A([Send Dolphin's answer]) C -- "below 75" --> P[" 3. Escalate Claude Opus 5.5 on SERV"] P --> B([Send the premium answer])

  1. Dolphin answers. DeepSeek V4.1 Flash on the Dolphin network, at $0.01 in and $0.07 out per million tokens.
  2. SERV checks. A small SERV Reasoning model scores the answer for correctness, whether it followed the instructions, and whether anything was made up.
  3. Escalate only if needed. Below 75, the question goes to Claude Opus 5.5 on SERV and that answer is returned instead.

Results

The benchmark runs every question two ways, Cheap First and premium only, and grades both final answers with the same checker. Every cost comes from the token counts the APIs returned.

Cheap FirstPremium only
Total cost, 143 questions$0.27$1.33
Saving80% (98% on 104 everyday questions, 66% on 39 stress tests)
Answered by Dolphin alone139 of 143
Average quality score99.798.0

The safety net works. The checker escalated 4 questions. Three were real catches: a wrong count, and blank answers on a 40-digit multiplication and the "world's hardest Sudoku". The premium model got all three right. The fourth was a false alarm on a correct answer, and the page shows it openly. In a separate test, the checker failed all 8 deliberately wrong answers it was given.

Harder questions

A fair criticism of the benchmark above is that the questions were too easy. So there is a hard benchmark: 97 hard, real-world style questions (multi-step maths, trick facts, strict format rules, failing tools, messy data, plus a "breaking point" round). Every answer is graded by code against a known correct answer, and the checker gets no answer key.

97 hard questionsCheap FirstPremium only
Correct, checked by code97 of 9797 of 97
Total cost$0.27$1.42
Saving81%

To test the checker on real mistakes, three weaker models answered the same questions: of 124 wrong answers, the checker caught 115 (93%). Of the 9 it missed, 5 were only a wrong status label and 4 were real errors, mostly facts the checker also got wrong. It raised false alarms on 13 of 166 right answers (8%). Full details, including the weak spots, are in docs/hard-benchmark.md.

Run it yourself

npm install

export DOLPHIN_API_KEY=... # from app.dphn.ai → API Keys export SERV_API_KEY=... # from console.openserv.ai npm run ask -- "How many seconds are there in 3.5 hours?"

It prints the answer, which model gave it, the check score and what it cost. In your own code:

import { cheapFirst, defaultModels } from "./src/cheapFirst";

const result = await cheapFirst("Summarise this email in one line: ...", await defaultModels()); console.log(result.answer, result.answeredBy, result.costUsd);

CommandWhat it does
npm run ask -- "question"Answer one question the Cheap First way
npm run benchmarkRun the benchmark and save data/results.json (pass question ids to run just those)
npm run hard-benchmarkRun the hard benchmark, graded by code (-- --set stress for round 2, see docs)
npm run buildBuild the results page into out/ (no keys needed)
npx tsx scripts/check-sanity.tsGrade deliberately wrong answers and save data/checker-test.json
npx tsx scripts/probe.ts file.jsonTry questions on Dolphin and the checker only, to find ones Dolphin gets wrong

Live demo server

api/ask.ts is a small serverless function (deployed on Vercel, see vercel.json) that powers the "Try it live" box on the results page. The keys stay on the server. Every call spends real credits, so it limits questions to 600 characters, 10 an hour per visitor and 300 a day per server (CF_PER_IP_PER_HOUR, CF_DAILY_CAP), and CF_MAX_TOKENS caps each model reply. The GitHub Pages build points that box at https://cheap-first.vercel.app/api/ask.

Spend guards (free demo and paid api/pay.ts)

An escalation to the premium model can cost far more than the $0.005 a paid answer earns (one benchmark escalation cost $0.19), so both endpoints:

  • cap each premium answer at CF_PREMIUM_MAX_TOKENS tokens (default 2000, so at most about $0.05),
  • stop escalating once a server instance has spent its daily premium budget (CF_PAID_PREMIUM_BUDGET_USD, default $2; CF_DEMO_PREMIUM_BUDGET_USD, default $1). A paid request that would need the premium model after that gets a 503 and is never charged,
  • go straight to the premium model when Dolphin errors out or returns a blank answer, instead of failing or paying to check an empty reply. The benchmark keeps outages visible and does not fall back.

The paid endpoint accepts questions of up to 2,000 characters, and an optional mode: strict (pass mark 85, escalates more often), balanced (75, the default) or budget (60). Each answer includes costUsd and savedUsd (the saving against the premium model alone; an estimate for cheap answers, and it can be negative for an escalation).

OpenAI-compatible API

api/v1/chat/completions.ts serves Cheap First in the OpenAI chat completions format, paid per answer with x402 like /api/pay. An app using an OpenAI-style client changes its base URL to https://cheap-first.vercel.app/api/v1 and wraps fetch with an x402 client. It supports messages (system, user, assistant, up to 6,000 characters in total), model (cheap-first, cheap-first-strict, cheap-first-budget) and stream (the answer arrives as a single chunk, since it is only released after the check). Each reply has a cheap_first field with who answered, the check score, the cost and the saving. GET /api/v1/models lists the model names.

MCP: Claude, Cursor, and other agents

Cheap First speaks the Model Context Protocol over Streamable HTTP, on the same deployment as the rest of the API:

https://cheap-first.vercel.app/mcp

https://cheap-first.vercel.app/api/mcp is the same endpoint. It is stateless: one JSON-RPC message per POST, no session.

ToolWhat you sendWhat you get back
ask_cheap_firstquestion, and an optional mode of strict, balanced (default) or budgetanswer, escalated, checkScore (0 to 100), costUsd, and savedUsd (the estimated saving against the premium model alone; it can be negative when an escalation costs more)
cheap_first_statsnothingThe published benchmark: 143 questions, $0.27 against $1.33, how many Dolphin answered, the quality scores, and the harder 97-question run

Free calls use the demo limits above: 600 characters, 10 an hour from one IP, 300 a day, and the demo's premium budget. They share those counters with /api/ask, so using both does not double the free tier. A prepaid key sent as Authorization: Bearer cf_... is charged one credit through the same path as /api/pay. An x402 payment of $0.005 USDC on Base works too. Send the header X-Cheap-First-Paid: 1 and the endpoint replies 402 with the usual challenge, then settles only after the answer is ready. The price is the same $0.005. Card payments stay off until Stripe is configured, which this server does not turn on.

Cursor

In ~/.cursor/mcp.json (everywhere) or .cursor/mcp.json in a project. Leave headers out for the free demo. ${env:CHEAP_FIRST_API_KEY} is read from the environment Cursor was launched with.

{

"mcpServers": { "cheap-first": { "url": "https://cheap-first.vercel.app/mcp", "headers": { "Authorization": "Bearer ${env:CHEAP_FIRST_API_KEY}" } } } }

Claude Desktop and Claude Code

In Claude Desktop: Settings → Connectors → Add custom connector, and paste https://cheap-first.vercel.app/mcp. Add the bearer header there if you have a prepaid key.

Claude Code, in .mcp.json:

{

"mcpServers": { "cheap-first": { "type": "http", "url": "https://cheap-first.vercel.app/mcp", "headers": { "Authorization": "Bearer ${CHEAP_FIRST_API_KEY}" } } } }

Clients that only launch a local process

src/mcpStdio.ts reads MCP messages on stdin and posts them to the hosted endpoint, so the demo limits and payments still happen on the server. From a checkout of this repo (npm install first):

{

"mcpServers": { "cheap-first": { "command": "npx", "args": ["tsx", "src/mcpStdio.ts"], "env": { "CHEAP_FIRST_API_KEY": "cf_..." } } } }

npm run mcp runs the same bridge. CHEAP_FIRST_MCP_URL overrides the endpoint. CHEAP_FIRST_PAID=1 sends the x402 header. The package is not on npm yet, so npx only works inside a checkout, where tsx is installed.

Listing in MCP directories

server.json is the metadata for the official MCP Registry. .mcp.json is the free demo: https://cheap-first.vercel.app/mcp and no headers. The registry entry can also offer an optional Authorization: Bearer cf_... prepaid key, and an optional X-Cheap-First-Paid: 1 header that asks for the $0.005 USDC on Base challenge. Leave both empty for the free demo. Its icon is https://cheap-first.vercel.app/icon.png, which the site build copies from brand/icon.png.

To publish a change, raise version in server.json and SERVER_VERSION in src/mcp.ts together (the registry refuses a version it already has), then run these yourself:

  1. Install mcp-publisher (brew install mcp-publisher, or the binary from the registry releases).
  2. Run mcp-publisher validate.
  3. Run mcp-publisher login github.
  4. Run `

只显示了开头部分;其余内容请在仓库中阅读