AI-First Scraper

io.github.yubinkim444由 GitHub 帳號 yubinkim444 發布v1.0.2更新於 Oct 25, 2026

Three MCP tools: fetch_page, fetch_pages_batch, search_web. Ad-free Markdown for AI agents.

發布方自己的服務:AI-first company: evidence on AI adoption

已驗證STDIO僅桌面網頁搜尋與抓取

概覽

AI 產生的概覽

一個本機 MCP 伺服器,為助理提供三個工具:把網頁擷取為乾淨的 Markdown,以及執行網頁搜尋並直接回傳已轉為 Markdown 的結果。

可以做什麼
擷取單一 URL,回傳乾淨的 Markdown,支援 HTML 與 PDF 頁面,去除廣告與原始標記。平行擷取最多 25 個 URL,並分別回傳每個頁面的 Markdown。執行網頁搜尋,回傳已轉為 Markdown 的熱門結果,可直接用於摘要或推理。
適用情境
適合助理需要閱讀網頁或 PDF、並進行網頁搜尋而不必處理原始 HTML 的情境,例如資料研究、文章摘要或為任務蒐集來源。
執行需求
以 stdio 在本機執行,可透過 uvx 從 PyPI 套件 ai-first-scraper-mcp 啟動,或以 pip 安裝。需要網路存取預設的公開擷取與搜尋服務,或透過 SCRAPER_URL 與 SEARCH_URL 指向自架後端。AFS_TIMEOUT 設定上游 HTTP 逾時。未宣告身分驗證。
安裝前請注意
預設情況下,所有擷取的 URL 與搜尋查詢都會送到託管於 Render 的公開 ai-first-scraper 與 ai-first-search 服務,頁面內容與查詢會離開本機。如有疑慮,可將 SCRAPER_URL 與 SEARCH_URL 指向自架後端。這些工具只讀取網頁內容,不涉及寫入或憑證。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 AI-First Scraper,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。

Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。

其他 MCP 客戶端

參照 儲存庫 中的啟動說明。

README

ai-first-scraper-mcp

Plug Claude Desktop, Cursor, or Cline straight into an ad-free web scraper + search engine. Three tools, one line of config.

[PyPI] [Python] [MCP] [License: MIT]


What it does

Adds three tools to any MCP-compatible agent:

ToolWhat it does
fetch_pageFetch one URL → return clean Markdown (HTML or PDF).
fetch_pages_batchFetch up to 25 URLs in parallel → return Markdown for each.
search_webRun a web search and return the top-k result pages already converted to Markdown.

No more "the model called curl and then tried to parse 80kB of ad HTML." Your agent receives clean Markdown ready to reason about.

Backed by the ai-first-scraper and ai-first-search APIs.


Install

Fastest — uvx (no install, runs from PyPI on demand)

jsonc
// claude_desktop_config.json  /  cline_mcp_settings.json  /  ~/.cursor/mcp.json{  "mcpServers": {    "ai-first-scraper": {      "command": "uvx",      "args": ["ai-first-scraper-mcp"]    }  }}

Restart your client (Claude Desktop / Cursor / Cline). The three tools above will appear automatically.

Alternative — pip install

bash
pip install ai-first-scraper-mcp
jsonc
{  "mcpServers": {    "ai-first-scraper": {      "command": "ai-first-scraper-mcp"    }  }}

Where the config file lives

ClientConfig path
Claude Desktop (macOS)~/Library/Application Support/Claude/claude_desktop_config.json
Claude Desktop (Windows)%APPDATA%\Claude\claude_desktop_config.json
Cursor~/.cursor/mcp.json
Cline (VS Code)~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json

Point at your own backend (optional)

By default this server calls the public ai-first-scraper.onrender.com and ai-first-search.onrender.com instances. If you want to self-host, set env vars in your MCP config:

jsonc
{  "mcpServers": {    "ai-first-scraper": {      "command": "uvx",      "args": ["ai-first-scraper-mcp"],      "env": {        "SCRAPER_URL": "https://your-scraper.example.com",        "SEARCH_URL":  "https://your-search.example.com",        "AFS_TIMEOUT": "60"      }    }  }}

Verify it works

Open your MCP client and ask the agent:

"Use the search_web tool to find the top 3 recent articles about MCP and summarize them in 5 bullets each."

You should see the agent call search_web, get back Markdown for each result, and produce the summary without ever touching raw HTML.


Companion projects

  • ai-first-scraper — the per-URL Markdown cleaner this MCP server fans out to.
  • ai-first-search — search → scrape → markdown pipeline.
  • mcp-rec — record & replay any MCP server's traffic for tests and bug reports.
  • llm-cache-proxy — local cache for OpenAI/Anthropic API calls.
  • promptlocker — lockfile for prompts.
  • context-diff — see what blew up your Claude Code context window.
  • agentwatch — overlay for browser AI agents.

Develop locally

bash
git clone https://github.com/yubinkim444/ai-first-scraper-mcp.gitcd ai-first-scraper-mcp
uv sync                    # or: pip install -e .ai-first-scraper-mcp       # speaks MCP over stdio

To test against a local client, point its MCP config at the same command.


About the author

Built by yubinkim444, who also makes Kay's Records — an app for iOS and Android.

If this project saved you time, giving the app a try is the nicest way to say thanks.

License

MIT © yubinkim444

來源:README.md,提交 ec64d75

工具

0
SourceWeft 尚未在沙箱中啟動過該軟體包,因此其工具未知。

版本歷史

1
  1. v1.0.2最新Sep 16, 2026

評分

已滿足適用檢查項的 50%

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25 分

    請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。

  • 已宣告認證方式10 分

    請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。

  • 已列出工具15 分

    請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。

  • 描述必需15 分
  • 圖示10 分
  • 文件15 分
  • 授權條款10 分

其他 Ai-First 整合