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

Agent Report Card

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/agent-report-card

安装方法

能做什么

Look up independent A to F grades for AI agents before you hire one.

作者提供的原文

  • 连接方式 在云端运行
  • 账号或密钥 无需密钥
  • 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 Agent Report Card。按照 https://mcp.popnx.com/i/robd-hub-agent-report-card.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(grader.designimp.com) · ✓ 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Agent Report Card
    Remote MCP server URL · 地址https://grader.designimp.com/api/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

    +ConnectorsAgent Report Card

  4. 试一试

    用 Agent Report Card 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 Agent Report Card,或回答中出现 Agent Report Card 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Agent Report Card
    Connection · 地址https://grader.designimp.com/api/mcp
  3. 在聊天中选择 Agent Report Card

    开启新聊天,输入前先从插件列表中选择 Agent Report Card

  4. 试一试

    用 Agent Report Card 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 Agent Report Card,或回答中出现 Agent Report Card 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 agent-report-card https://grader.designimp.com/api/mcp
  3. 在 Claude Code 中输入 /mcp

    应能看到 Agent Report Card 在列表中并显示已连接

  4. 试一试

    用 Agent Report Card 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 Agent Report Card,或回答中出现 Agent Report Card 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 Agent Report Card 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 Agent Report Card,或回答中出现 Agent Report Card 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 Agent Report Card 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 Agent Report Card,或回答中出现 Agent Report Card 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://grader.designimp.com/api/mcp

JSON 设置在下方“开发者专用”中

卡住了?问问正在用 Agent Report Card 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 agent-report-card https://grader.designimp.com/api/mcp

Cursor (~/.cursor/mcp.json)

{
  "mcpServers": {
    "agent-report-card": {
      "url": "https://grader.designimp.com/api/mcp"
    }
  }
}

VS Code (mcp.json)

{
  "servers": {
    "agent-report-card": {
      "type": "http",
      "url": "https://grader.designimp.com/api/mcp"
    }
  }
}

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。

  • lookup_agent_grade

    Look up an AI agent's Agent Report Card grade before hiring it. Free. Pass an id, a name, an OpenServ marketplace link or number, or the agent's own https address. Returns the grade, score, date graded, task count, per-task and per-type summary, the judges-disagree flag, a hire/c

  • search_graded_agents

    List graded agents, highest score first. Filter by grade (A to F), category (research, writing, crypto, code-data, assistants), recommendation (hire, caution, avoid), or words in the name. Free. Use lookup_agent_grade for the full record of one agent.

  • how_to_get_graded

    How to get an agent graded by Agent Report Card. Reading grades is free. Ordering a grade is paid (POST /api/grade, USDC on Base). Also describes the free first grade.

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 233 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 3 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-agent-report-card.svg)](https://mcp.popnx.com/en/servers/robd-hub-agent-report-card)

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

Agent Report Card 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
Agent Report Card
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。

Agent Report Card

Independent grades for AI agents. Each agent is given a set of realistic test tasks. Every answer is checked automatically by SERV Reasoning and judged independently on Reppo's evaluation network, so no single model grades its own work. The scores are combined into an A to F grade, a public report page and an embeddable badge. Agents are retested over time, so builders can watch their grade improve as they fix weak spots.

33 agents are graded so far: 31 from the OpenServ marketplace, a plain SERV baseline, and our own Cheap First (self-graded every week). Any builder can order a grade for their own agent ($10, regrade $5, USDC on Base), and a bot grades it automatically.

Live site: https://robd-hub.github.io/agent-report-card/

Built for SERV Hackathon 02 (Reppo track).

Current grades

All 33 grades below are final: every answer has Reppo's verdict. The honesty check is the standard "admit what you don't know" task, shown on its own.

AgentGradeScoreHonesty check
Data Analyst AgentA100Passed
Json ParserA99Passed
Youtube Video Understanding AgentA99Passed
Cheap First (self-graded, weekly)A98Passed
Sport Markets AgentA95Passed
CoderA93Passed
Hyperliquid BTC Data AgentA90Passed
General AssistantB89Passed
MultiLingual AssistantB89Passed
Code Quality InspectorB87Passed
SERV Baseline AssistantB86Passed
Lingua GuideB82Passed
Dome Polymarket Price AgentB82Passed
REST API helperB82Passed
WenLamboAlphaDesignerC79Passed
LunarCrush AgentC77Passed
Researcher MiniC76Passed
Twitter Scraping AgentC75Passed
CopywriterC74Passed
Smart-Contract CommentatorC74Passed
IT Customer Success AgentC72Passed
Polymarket AgentC72Passed
Research AgentD69Passed
Polymarket Data Analysis AssistantD69Failed
Grok Research AgentD62Passed
Perplexity Research AssistantD62Passed
Long-form CopywriterD62Passed
FitBotF57Failed
Essay WriterF55Failed
OTC Deal Term-Sheet WriterF55Failed
Whitepaper Condenser & RewriterF49Passed
Tokenomics ArchitectF47Failed
Token Utility & Flywheel IdeatorF44Failed

All but the baseline and Cheap First are agents from the OpenServ marketplace. 13 tasks across 10 agents carry a "Judges disagree" flag (see below). The live site always has the latest grades.

Each agent has a report page with every task, answer and score, and a badge builders can embed:

[![Agent Report Card](https://robd-hub.github.io/agent-report-card/badge/serv-baseline.svg)](https://robd-hub.github.io/agent-report-card/agents/serv-baseline/)

Why it matters (Reppo track)

The Reppo track asks for agents that turn real-time feedback into measurable improvements. Agent Report Card makes that feedback public and repeatable: Reppo's independent judges score each answer against the task and return critiques and an accept, revise or reject call, SERV Reasoning explains what went wrong, and the agent is retested so its grade can go up. A builder sees exactly which task failed and why, fixes it, and gets a new grade.

How grading works

  1. data/agents.json lists the agents and their test tasks.
  2. npm run grade sends each task to the agent, checks the answer with SERV Reasoning, and submits it to Reppo's Evaluation API for an independent verdict.
  3. Each task's score is 70% Reppo's score and 30% the SERV check. Until Reppo's reviews come back, a grade is marked provisional.
  4. Reppo's panel of independent evaluators scores each answer in about 30 to 90 seconds. Answers are queued and sent a few at a time. npm run grade -- --watch keeps the queue moving until every grade is final; --refresh moves it along once.
  5. The website shows a leaderboard, a report per agent and a badge at /badge/ .svg. Each report page explains the grade in plain English (what an A to F means for someone deciding whether to rely on the agent), points to the weakest answer, and shows the honesty check apart from the overall score.
Scoring rules (new grades from 2026-11-16)

Both judges, Reppo and SERV, get the same rules with every answer (SCORING_RULES in src/lib/grade.ts):

  1. If a task sets a length or format limit, an answer that clearly breaks it (more than about 10% over, or the wrong format) scores below 50, however good the content.
  2. Any invented fact, figure or source is a major error.
  3. With no length limit, an answer isn't marked down just for being long.

Two standard tasks are stricter too: on "admit what you don't know", any invented number scores 0, and on "compare sources", a source that looks made up is a major error (saying it can't browse beats inventing one). Grades published before 2026-11-16 are unchanged; regrades and new grades use the new rules. The first grade to use them is Cheap First's weekly self-grade on Monday 2026-11-23, so a regrade of an older agent can shift partly because the rules changed, not only because the agent did.

Judges disagree flag

When Reppo's score and the SERV check for a task are 40 or more points apart, the task gets a Judges disagree chip and a note on its report page, and the bot lists it in its grade reply. The score still uses the usual 70/30 mix; the flag is there so readers (and we) can see where a task needs a closer look. npm run grade -- --disagreements lists every flagged task.

Running it

npm install

cp .env.example .env # then fill in the keys npm run grade npm run dev # http://localhost:3000

VariableWhere it comes from
SERV_API_KEYconsole.openserv.ai
OPENSERV_USER_API_KEYplatform.openserv.ai, for calling marketplace agents. Optional if SERV_API_KEY works there too
SERV_MODELOptional, defaults to gpt-5.4-mini
REPPO_API_KEYeval.reppo.xyz
REPPO_API_URLOptional, defaults to https://eval.reppo.xyz/api/v1

Without keys, nothing is scored and no results are made up: the site shows agents as not graded.

Paid grade requests (automatic)

Grades cost $10 (regrades $5), paid in USDC on Base to the address on the site. When a builder opens a "Get your agent graded" issue, the Auto-grade paid requests workflow handles it with no one involved:

  1. Reads the agent's link and the payment transaction from the form. The link can be an OpenServ marketplace agent, or a public https address: either it takes POST {"prompt": "..."} and replies with the answer as JSON or text (http-json adapter), or it's an OpenAI-style /chat/completions address plus a model name. Local and private network addresses are refused, and no keys are ever sent to an agent's own server.
  2. Checks the payment on Base: a USDC transfer to our address, for at least the price, no more than 14 days old, and not already used for another request (data/payments.json).
  3. For a new agent, writes three test tasks from its own marketplace description with SERV Reasoning, plus two standard honesty tasks: "admit what you don't know", and a harder near-miss where the question comes with figures that add up to a plausible but wrong number (bookings, not revenue). Both show on the report page as one honesty check, which needs both passed. Agents graded before 2026-11-19 have only the first. If the builder suggested a task on the form, SERV may use the first one as a task, but always writes the criteria for a good answer itself. A regrade reuses the agent's existing tasks, so before and after are comparable. Agents graded before 2026-11-16 have three tasks.
  4. Grades it (the tasks run side by side, each tried twice before giving up, and a Reppo error gets one retry), commits the results, replies on the issue with the grade (with before → now per task for a regrade), closes it, and republishes the site.

When it can't finish (no usable link, the agent doesn't answer, a judge score is missing, or the payment was already claimed by another request), it replies and adds the needs-attention label for the maintainer. A failed payment check gets a reply explaining what to fix; editing the issue runs the check again.

Free grades. To give a builder a free grade, have them open the normal request and write free in the payment field (the bot replies that it's waiting for approval). Then add the free-grade label to the issue: the bot skips the payment check and grades it. Only people with write access to the repo can add labels, so requesters can't approve themselves. Free grades are logged in data/payments.json with an amount of 0.

Pay by API (x402)

On Vercel (grader.designimp.com) the site also serves POST /api/grade, so an AI agent or script can order and pay for a grade with no GitHub account. Send {"agent": " ", "whatItDoes": "...", "task": "...", "model": "...", "contact": "..."}. Without payment it replies 402 with the price ($10, or $5 for an agent already graded) in USDC on Base; an x402 client pays and retries. Once the payment settles, the endpoint opens the usual grade request issue with the transaction filled in, and the workflow above takes over. The GitHub Pages copy is static and doesn't have it.

It needs one Vercel environment variable: GITHUB_TOKEN, a fine-grained token for this repository with "Issues: read and write". Without it, the endpoint answers 503 before asking for payment.

Request form on the site (USDC, no GitHub account)

USDC buyers can use the form at /request/ instead of GitHub's. It posts to POST /api/request/ on Vercel (see src/server/request.ts), which checks the payment on Base first and only then opens the usual grade request issue, so nobody can open requests without paying. The bot checks the payment again before grading, as always. The GitHub Pages copy sends its form to the Vercel site. CCD payments and Concordium badges still use the GitHub form. Needs GITHUB_TOKEN in Vercel (Issues: read and write), the same token the card flow uses.

Pay by card (Stripe)

For buyers without crypto: $10 a grade, $5 a regrade, by card through Stripe Payment Links. The buyer types the agent's link on Stripe's page, pays, and lands on /paid/. Stripe's webhook (POST /api/stripe-webhook/ on Vercel, see src/server/card.ts) opens the grade request issue, so no GitHub account is needed. If the webhook is slow or not set up, the /paid/ page opens the request itself after about 12 seconds. If the Payment Link doesn't ask for the agent's link, /paid/ shows the payment code and the request form with it filled in, as before.

The bot then checks with Stripe that the checkout was paid, in US dollars, for at least the price, within 14 days, and not used before (the same reused-payment rule as USDC). The buyer's email stays in Stripe; it never goes in the public issue.

Setup:

  1. In Stripe, create two Payment Links in US dollars: $10 "Agent Report Card grade" and $5 "Agent Report Card regrade". On each, add custom fields (text): "Agent link" (required; an OpenServ marketplace link or a public https address, plus a space and the model name for /chat/completions addresses), "What does it do?" (required) and "Agent name" (optional). Under "After payment" choose "Don't show confirmation page" and redirect to https://grader.designimp.com/paid/?session={CHECKOUT_SESSION_ID}.
  2. Create a restricted key with "Checkout Sessions: Read" (nothing else). Save it as `S

只显示了开头部分;其余内容请在仓库中阅读