Webfetch.io

io.webfetch由 webfetch.io 發布官方發布者v1.0.0更新於 Nov 22, 2026

Fetch any URL as markdown, text, HTML or selector-built JSON; search Google, Bing, Yahoo and Brave.

已驗證Streamable HTTP可網頁執行網頁搜尋與抓取瀏覽器自動化

概覽

AI 產生的概覽

遠端 MCP 伺服器,讓助理把任意網址抓取為 Markdown、純文字、HTML 或以選擇器組出的 JSON,並可搜尋 Google、Bing、Yahoo 和 Brave。

可以做什麼
  • 把網址抓取為 Markdown、純文字或原始 HTML,預設會渲染 JavaScript
  • 用 CSS 或 XPath 選擇器從頁面擷取指定欄位並以 JSON 回傳
  • 搜尋 Google、Bing、Yahoo 或 Brave,以 JSON 回傳自然結果、廣告、相關搜尋和問題
適用情境
適合助理需要即時網頁內容的情況:讀取已知網址的頁面、從頁面擷取價格或表格等特定欄位,或在還沒有網址時先搜尋找到頁面。
執行需求
遠端 Streamable HTTP 端點,不需在本機安裝套件。每次請求都必須以 Authorization Bearer 權杖形式提供 API 金鑰;免費金鑰初始 1,000 點,每月再送 250 點。搜尋功能需在金鑰上開通搜尋權限,這需要付費方案並開啟 robots.txt 退出選項。
安裝前請注意
呼叫以點數計費:不渲染 JavaScript 的抓取 1 點,渲染為 5 點,搜尋每頁 10 點(light_request 為 false 時 15 點);失敗的呼叫不計費。金鑰以 Bearer 權杖隨每次請求傳送,需妥善保管。搜尋權限取決於金鑰上的 robots.txt 退出設定。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Webfetch.io,將其新增到工作區。
  2. 連接 Authorization token 憑證。它們會被私密儲存,絕不會在公開頁面上展示。
  3. 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/io.webfetch%2Fwebfetch/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": {
    "webfetch-io": {
      "type": "http",
      "url": "https://webfetch.io/mcp",
      "headers": {
        "Authorization": "Bearer <YOUR_TOKEN>"
      }
    }
  }
}

README

Webfetch SDKs

Official clients for the webfetch.io scraping and search API, in JavaScript, Python and PHP. MIT licensed.

Each one wraps GET https://api.webfetch.io/v1/ and the search endpoints (/v1/google, /v1/bing, /v1/yahoo, /v1/brave). The API is ScrapingBee-compatible, so parameter names, defaults and credit costs match.

LanguageInstallRequires
JavaScriptnpm install @tuxxin/webfetchNode 18+, no runtime dependencies
Pythonpip install webfetch-io (imports as webfetch)Python 3.9+, httpx
PHPcomposer require tuxxin/webfetchPHP 8.1+, ext-curl

The shape is the same in all three

client = Webfetch(api_key)result = client.fetch(url, render_js=True, ...)
result.content     the page, as bytes or a stringresult.credits     what this call cost, from the Wf-Cost headerresult.status      the status the target returnedresult.resolved_url  where it ended up after redirects

Search

results = client.bing("python asyncio tutorial", country_code="us")results["organic_results"]   [{position, title, url, displayed_url, description, ...}]results["top_ads"]           [{position, title, url, tracking_url, domain, ...}]
client.search("yahoo", query, search_type="ads")   the ads only, same price

The response has the shape of ScrapingBee's Google Search API. Needs a paid plan with the robots.txt opt-out turned on for the key in the key portal: every engine's robots.txt disallows its result pages. Google, Bing, Yahoo and Brave are all open. 10 credits a page, 15 with light_request false; a failed search costs nothing. Reference: https://webfetch.io/documentation#search.

Credits

RequestCredits
No JavaScript rendering1
JavaScript rendering (the default)5
Premium proxy10, or 25 with rendering
Stealth proxy75
Search, light request (the default)10
Search, light_request false15
Any failed request0

mode=auto tries those routes cheapest first and charges only the one that worked; max_cost caps it. An anti-bot challenge or CAPTCHA page is never charged.

render_js defaults to true, so a call that does not set it costs 5 credits. Pass render_js=false for a plain HTTP fetch at 1 credit.

A key is required

There is no keyless tier. A free account starts with 1,000 credits and gets 250 more each month (banking up to 1,000), and needs only a confirmed email address. Each client raises WebfetchError immediately when constructed without a key, rather than spending a round trip on a request that cannot succeed.

Errors

Every SDK raises the same three-level hierarchy, so you can catch broadly or narrowly:

WebfetchError            anything went wrong  WebfetchApiError       the API answered with an error status; carries .status and .code    WebfetchRateLimitError   429, or 503 when free capacity is busy; carries .retry_after

A failed call is never billed, so you do not need to reconcile credits after an error.

MCP server

Webfetch is also a remote MCP server, so an AI assistant can fetch pages and run searches as tools, with no SDK in between.

Endpointhttps://webfetch.io/mcp (Streamable HTTP, POST only)
Registryio.webfetch/webfetch in the official MCP Registry; manifest: server.json
AuthAuthorization: Bearer wfk_... on every request. A free key starts with 1,000 credits and gets 250 more each month
Docshttps://webfetch.io/documentation#mcp
ToolWhat it does
fetch_pageA URL as markdown, plain text or raw HTML. JavaScript is rendered by default (5 credits); render_js false is a plain fetch (1 credit)
extract_dataJSON built from CSS or XPath selectors, so the model gets fields instead of a whole page
search_webOne page of Google, Bing, Yahoo or Brave results as JSON: organic results, ads, related searches and questions. 10 credits a page; needs search access on the key

A failed call costs nothing. Claude Code, in .mcp.json:

json
{  "mcpServers": {    "webfetch": {      "type": "http",      "url": "https://webfetch.io/mcp",      "headers": { "Authorization": "Bearer ${WEBFETCH_KEY}" }    }  }}

Documentation

https://webfetch.io/documentation

來源:README.md,提交 d5ce09c

工具

3

fetch_page

Fetch a web page

Fetch one URL and return the page as markdown, plain text or raw HTML. Use it to read a page whose URL you have; use extract_data instead for specific fields as JSON, and search_web when you have no URL yet. JavaScript is rendered by default (5 credits); render_js false is a plain HTTP fetch (1 credit). The result is a summary line (host, credits charged) and then the content, cut at 100,000 characters, which the summary notes. A failed fetch returns isError with the HTTP status and reason, and costs nothing.

輸入參數url必填render_jsreturn_format

extract_data

Extract structured data from a page

Fetch one URL and return JSON built from CSS or XPath selectors: named fields instead of a whole page. Use it when you know which elements hold the data (prices, titles, links, tables); read the page with fetch_page first if you do not. It costs what fetch_page costs. A rule that matches nothing returns null (item) or [] (list) and the call still succeeds, so check for nulls. An invalid rule or a failed fetch returns isError with the reason, and costs nothing. Output is cut at 100,000 characters.

輸入參數url必填render_jsextract_rules必填

search_web

Search the web

Search Google, Bing, Yahoo and Brave (one engine per call) and return one results page as JSON: organic_results (position, title, url, description, date), top_ads, bottom_ads, related_searches, questions and number_of_results. Use it to find pages when you have no URL; then read a result with fetch_page or pull fields from it with extract_data. 10 credits per page. Needs search access on the key: paid plans turn on the robots.txt opt-out in the key portal, and free accounts request it there. Without it, or on any failure, the call returns isError with the reason, and costs nothing.

輸入參數pagequery必填enginesearch_typecountry_codelight_request

版本歷史

1
  1. v1.0.0最新Nov 22, 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

其他 Webfetch 整合