AgenticWebScore

com.agenticwebscore由 agenticwebscore.com 發布官方發布者v0.3.0更新於 Oct 29, 2026

Scan a public website for AI agent readiness: 60 checks, a score, and the evidence for each.

已驗證Streamable HTTP可網頁執行網頁搜尋與抓取開發者工具

概覽

AI 產生的概覽

遠端 MCP 伺服器，掃描公開網站的 AI 代理就緒度，回傳評分、60 項檢查結果與證據。

可以做什麼
  • 掃描公開網域並回傳完整紀錄：評分及其兩部分、等級、每項檢查狀態、證據與指紋
  • 讀取某主機已儲存的掃描紀錄，不啟動新掃描
  • 取得某主機的 Markdown 狀態報告，或依檢查 ID 取得單一檢查
  • 列出 rubric v0.9 的 60 項檢查及其類型、模組、維度、理由與計分方式
適用情境
適合評估公開網站在多大程度上服務 AI 爬蟲與代理（robots.txt 與 AI 爬蟲規則、網站地圖、結構化資料、快取、llms.txt、OpenAPI 與 MCP 伺服器），或說明既有評分及未通過的檢查項目。
執行需求
需要可連線的遠端 Streamable HTTP 端點；不需 API 金鑰、註冊或本機執行環境。用戶端須支援 HTTP 傳輸及所列協定版本之一，並能連線至託管端點與被掃描網站。
安裝前請注意
scan_site 並非唯讀且屬開放世界：會向指定網域發出對外 GET 請求，若該網站執行 MCP 伺服器還會進行一次 MCP 握手（但不會呼叫其工具）。新掃描限制為每用戶端每分鐘 10 次、同時 3 個，單次最長約 60 秒。被掃描的網域會傳送至託管服務。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 AgenticWebScore，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

交給代理程式安裝

把這段提示詞貼上到 Claude Code、Codex、Cursor 等代理程式，或 SourceWeft 對話中。代理程式會讀取該伺服器的安裝指南，先向你展示發布者、端點和登入方式，經你同意後再把它新增到你的客戶端。金鑰不會寫進任何命令。

閱讀 https://sourceweft.com/mcp/com.agenticwebscore%2Fscan/install.md，並按其中說明把這個 MCP 伺服器接入我的客戶端。修改任何設定之前，先告訴我它的發布者、端點和登入方式，等我確認。不要把 API 金鑰寫進命令或我會提交的檔案；問我金鑰應從哪裡取得。

檢視代理程式所依據的安裝指南

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "agenticwebscore": {
      "type": "http",
      "url": "https://agenticwebscore.com/mcp"
    }
  }
}

README

[AgenticWebScore]

AgenticWebScore MCP server

Scan a public website for AI agent readiness: 60 checks, a score, and the evidence for each.

AgenticWebScore measures how well a public website serves AI agents. One scan runs 60 checks from the outside, the way a crawler or an agent reaches the site:

  • robots.txt and AI crawler rules;
  • sitemaps, headings and structured data;
  • HTTP caching and real 404s;
  • markdown negotiation and llms.txt;
  • OpenAPI and MCP servers.

Each check returns its status, the evidence it read, the source it follows, and a shell command that repeats the measurement. The score is 70% adopted web standards and 30% agent conventions, and every record carries a fingerprint you can verify.

This repository holds the listing files for the hosted server: this README, server.json for the official MCP Registry, the logo and client configs. The server itself runs at https://agenticwebscore.com/mcp.

Connect

  • Transport: Streamable HTTP. The server is stateless: one JSON-RPC message per POST, no session, and GET answers 405.
  • Authentication: none. No API key and no sign-up.
  • Protocol versions: 2025-11-25 and 2025-06-18.

Claude Code:

sh
claude mcp add --transport http agenticwebscore https://agenticwebscore.com/mcp

Any client that reads an mcpServers file:

json
{  "mcpServers": {    "agenticwebscore": {      "type": "http",      "url": "https://agenticwebscore.com/mcp"    }  }}

Gemini CLI, as an extension (gemini-extension.json):

sh
gemini extensions install https://github.com/marifeting/agenticwebscore-mcp

This repository is also a Cursor plugin (.cursor-plugin/plugin.json and mcp.json) and carries a LobeHub manifest (lhm.plugin.json). All of them point to the same remote endpoint.

Tools

ToolWhat it doesAnnotations
scan_siteScans a public domain and returns the full record: the score and its two parts, the level, each check's status and evidence, and the fingerprint. A stored record younger than 6 hours is returned without a new scan, unless fresh is true.not read-only, open world
get_scoreReturns the stored record for a host. Starts no scan.read-only
list_checksLists the 60 checks of rubric v0.9 with their kind, module, dimension and rationale.read-only
get_reportReturns the markdown report of a stored record, or one check of it. Starts no scan.read-only

Example prompts

  • "Scan example.com for agent readiness and list the checks that failed."
  • "What does example.com's robots.txt say to GPTBot and ClaudeBot?"
  • "Read the stored AgenticWebScore report for cloudflare.com and explain its MCP result."
  • "Which of the 60 checks are about llms.txt, and how is each scored?"

Limits

  • A new scan counts against 10 per client per minute and 3 running at once.
  • A scan runs for up to 60 seconds. If a call times out first, the scan still finishes and is stored.
  • A scan sends only GET requests to the site, apart from an MCP handshake (initialize and tools/list) when the site runs an MCP server. It never calls that server's tools.

Links

Operated by Marifeting. Scanning is free.

License

The files in this repository are MIT-licensed (see LICENSE). The hosted service is covered by its terms.

來源：README.md，提交 d29d5e0

工具

4

scan_site

Scan a site

Scan a public domain and return the full scan record: score and its two parts, level, per-check status, evidence, and fingerprint. A stored record younger than 6 hours is returned without a new scan, unless fresh is true. New scans are limited to 10 per client per minute and 3 running at once. A new scan runs for up to 60 s. If the call times out first, the scan still finishes and is stored, and a later call returns it.

輸入參數freshtarget必填

get_score

Read a stored score

Return the stored scan record for a host. It starts no scan. A host with no stored record is answered with an error that says so.

輸入參數host必填

list_checks

List the checks

List the 60 checks of rubric v0.9 across 5 dimensions, each with its kind, module, dimension, and rationale. A scored check also has a part, a tier and a weight. An unscored check earns 0 points.

無需輸入參數。

get_report

Read a report

Return the markdown status report of a host's stored scan record, or one check of it by check id. It starts no scan.

輸入參數host必填check

版本歷史

1
  1. v0.3.0最新Oct 29, 2026

評分

已滿足適用檢查項的 100%

該評分衡量此收錄條目的完整程度，依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價，也不表示伺服器安全或目前可用：可連通性是帶日期的檢查結果，評分也不包含安裝量或使用者評價。

  • 近 30 天內可連通必需25 分
  • 已宣告認證方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分