AgentScrape

io.github.hshintelligence由 GitHub 帳號 hshintelligence 發布v0.6.1更新於 Oct 24, 2026

Pay-per-call web scraping for AI agents via x402 on Base USDC. Six tools, no signup.

發布方自己的服務：Scrape.ly

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，可擷取網頁、抽取結構化資料與中繼資料、截圖並執行瀏覽器工作流程，依次以 USDC 付費。

可以做什麼
  • 將網頁擷取為 markdown、HTML、文字或 JSON
  • 以自然語言提示或 schema 從頁面抽取結構化 JSON
  • 擷取 PNG 截圖，支援桌機、手機、平板視窗及整頁模式
  • 抽取頁面中繼資料：標題、描述、Open Graph、Twitter 卡片、JSON-LD 與標準連結
  • 建立保留 cookie 與 localStorage 的有狀態瀏覽器工作階段
  • 執行最多 20 步的多步瀏覽器工作流程（導覽、點擊、輸入、等待、捲動、截圖、抽取、執行腳本）
適用情境
適合需要即時網頁內容、結構化抽取或截圖、又不想自行建置擷取架構的助理，且用戶端能以加密錢包完成 x402 付款。也適合無法註冊帳號或管理 API 金鑰的自主任務代理。
執行需求
遠端 Streamable HTTP 端點，無需註冊或 API 金鑰。每次工具呼叫須透過 x402 協定以 Base 主網 USDC 付款，因此 MCP 用戶端需具備相容 x402 的錢包。每錢包每 30 天 10 次的免費額度僅適用於 HTTP API，不適用於 MCP 呼叫。
安裝前請注意
每次 MCP 工具呼叫都會從錢包扣除 USDC，價格在啟動期間標為 0.001 美元，之後調漲。付款由用戶端處理，錢包憑證與支出上限由使用者自行控管。擷取的網址與頁面內容會傳送給營運方及第三方服務（Cloudflare Browser Rendering、Groq）處理。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 AgentScrape，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/io.github.hshintelligence%2Fagent-scrape/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": {
    "agentscrape": {
      "type": "http",
      "url": "https://agent-scrape.healingsunhaven.workers.dev/mcp"
    }
  }
}

README

AgentScrape

Pay-per-call web scraping for AI agents — no signup, no API keys, just USDC.

[x402] [MCP] [Base] [Cloudflare Workers] [License: MIT] [hshintelligence/agent-scrape MCP server]

Live: agent-scrape.healingsunhaven.workers.dev

AI agents discover AgentScrape, pay per call in USDC on Base, and get clean structured data back — no accounts, no API keys, no human in the loop.

Two protocols, one service:

  • HTTP API with x402 v2 payment gate — for agents using raw HTTP
  • MCP server (Streamable HTTP transport) — for Claude Desktop, Cursor, Continue.dev, and any MCP-compatible framework

Why this exists

Traditional scraping APIs assume a human:

  • Sign up with email
  • Add a payment method
  • Manage an API key
  • Commit to a monthly subscription

AI agents can't do any of that. They have wallets. They have stablecoins. They need infrastructure built for them.

AgentScrape speaks two open standards:

  • x402 — HTTP-native payment protocol. The server returns 402 with payment requirements; the agent signs a USDC transfer; the facilitator settles on-chain; the data flows back. End-to-end in under 2 seconds.
  • MCP — the standard agent tool interface. Agents browse, discover, and invoke tools without per-vendor SDKs.

No signup. No API keys. No subscription. Just USDC.

Tools

All tools cost $0.001 USDC during the mainnet validation window. Pricing ramps after stability is proven (see Pricing).

ToolDescription
scrape_webpageScrape any URL to markdown, HTML, text, or JSON
extract_structured_dataAI-powered structured extraction with natural-language prompts and JSON schemas (Groq + Llama 4 Scout)
screenshot_webpagePNG screenshot with viewport control (desktop/mobile/tablet, optional full-page)
extract_metadataTitle, description, Open Graph, Twitter cards, JSON-LD, canonical URL
create_browser_sessionStateful browser session with cookie + localStorage persistence
run_workflowMulti-step atomic execution: navigate, click, type, wait, scroll, screenshot, extract, evaluate

Quick start — HTTP API

bash
# Free service discovery (no payment)curl https://agent-scrape.healingsunhaven.workers.dev/
# Paid call — without payment, returns 402 with payment requirementscurl -X POST https://agent-scrape.healingsunhaven.workers.dev/scrape \  -H "Content-Type: application/json" \  -d '{"url": "https://example.com", "format": "markdown"}'
# Free tier — first 10 calls per wallet, send your wallet in the X-402-Payer headercurl -X POST https://agent-scrape.healingsunhaven.workers.dev/scrape \  -H "Content-Type: application/json" \  -H "X-402-Payer: 0xYourBaseWalletAddress" \  -d '{"url": "https://example.com", "format": "markdown"}'

Quick start — MCP

Add to any MCP-compatible client via Streamable HTTP:

json
{  "mcpServers": {    "agent-scrape": {      "url": "https://agent-scrape.healingsunhaven.workers.dev/mcp",      "transport": "streamable-http"    }  }}

The x402 payment metadata is broadcast inside the MCP tools/list response. Your client handles the payment handshake automatically when configured with an x402-compatible wallet.

Discovery

For machine-driven discovery, four standard endpoints are exposed (free, no auth):

EndpointPurpose
/Service profile JSON (tools, prices, network, payTo)
/.well-known/x402 (and .json alias)x402 manifest — payTo, network, facilitator, all paid routes
/openapi.jsonOpenAPI 3.1 spec for traditional tooling
/llms.txtAgent-friendly plaintext description

The 402 response itself carries the canonical payment requirements in the base64-encoded Payment-Required header, and the body includes a human-readable error with discovery links.

Architecture

                ┌──────────────────────────────────┐                │      Cloudflare Workers          │                │      (global edge, 330+ POPs)    │                │                                  │AI Agent ───────┼─▶  GET  /                  (free)│(HTTP)          ├─▶  GET  /.well-known/x402  (free)│                ├─▶  GET  /openapi.json      (free)│                ├─▶  GET  /llms.txt          (free)│                ├─▶  POST /scrape         ($0.001) │                ├─▶  POST /extract        ($0.001) │                ├─▶  POST /screenshot     ($0.001) │                ├─▶  POST /metadata       ($0.001) │                ├─▶  POST /workflow       ($0.001) │                ├─▶  POST /session        ($0.001) │                │                                  │AI Agent ───────┼─▶  POST /mcp                     │(MCP)           │      • initialize                │                │      • tools/list                │                │      • tools/call (paid)         │                │                                  │                └──────────────┬───────────────────┘                               │              ┌────────────────┴─────────────────┐              │                                  │   Coinbase CDP facilitator       Browser Rendering API   (api.cdp.coinbase.com/         Groq API (Llama 4 Scout)    platform/v2/x402)             KV (sessions, free tier)   Ed25519 JWT auth   Settles USDC on Base              │              ▼   payTo wallet 0x3F33...9E9a

Stack:

  • Runtime: Cloudflare Workers (V8 isolates, global edge)
  • HTTP framework: Hono + @x402/hono v2 payment middleware
  • MCP framework: @modelcontextprotocol/sdk + agents/mcp with withX402 payment binding
  • Bazaar discovery: @x402/extensions — every paid route declares input schema + output example
  • Browser: Cloudflare Browser Rendering API + @cloudflare/puppeteer
  • AI extraction: Groq + Llama 4 Scout (17B, 16E)
  • Storage: Cloudflare KV (5-min response cache, browser sessions, free-tier counters)
  • Network: Base mainnet (eip155:8453)
  • Asset: Native USDC (0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913)
  • Facilitator: Coinbase CDP (api.cdp.coinbase.com/platform/v2/x402) with Ed25519 JWT auth

Free tier

Every wallet address gets 10 free calls per 30 days on the HTTP API. Pass your Base wallet in the X-402-Payer header. After 10 calls, payment via x402 is required.

Free tier is HTTP-only. MCP calls go through the standard x402 payment flow.

Pricing

Launch window (first 48h on mainnet): $0.001 flat per tool call.

After validation, pricing ramps to:

ToolProduction price
scrape_webpage$0.003
extract_structured_data$0.005
screenshot_webpage$0.003
extract_metadata$0.002
run_workflow$0.008
create_browser_session$0.001

Bulk discounts (20% at 1K calls/month, 40% at 10K) are planned.

Why x402 + MCP

Traditional APIx402 + MCP
Sign up at scraperapi.comDiscover via MCP initialize
Add credit cardWallet holds USDC
Get API keyNo keys
$49/month minimum$0.001 per call
Manage rate limitsPay-per-use, no limits
Human in the loopAgent fully autonomous

Verifying the deployment

bash
# Service identitycurl https://agent-scrape.healingsunhaven.workers.dev/ | jq .
# x402 manifestcurl https://agent-scrape.healingsunhaven.workers.dev/.well-known/x402 | jq .
# 402 challenge with x402 v2 headerscurl -i -X POST https://agent-scrape.healingsunhaven.workers.dev/scrape \  -H "Content-Type: application/json" \  -d '{"url":"https://example.com"}'# → HTTP 402, Payment-Required header (base64 JSON), Cache-Control: private, no-store
# CORS preflight (browser-based agents)curl -i -X OPTIONS https://agent-scrape.healingsunhaven.workers.dev/scrape \  -H "Origin: https://example.com" \  -H "Access-Control-Request-Method: POST" \  -H "Access-Control-Request-Headers: Content-Type,X-Payment"# → HTTP 204, allow-headers includes X-Payment and Payment-Signature

Roadmap

Q3 2026:

  • AgentParse — document/PDF/OCR-to-markdown service (same x402 + MCP pattern)
  • AgentSearch — multi-backend web search aggregation
  • LangChain integration package (langchain-agent-scrape)

Q4 2026:

  • Subscription bundles (1000 scrapes for $2.50)
  • Multi-asset support (USDT on Base, USDC on Arbitrum)
  • Webhook + async response option for long-running workflows

Run locally

Requires Node 22+, a Cloudflare account with Workers + Browser Rendering enabled, a Groq API key, and a Coinbase CDP API key.

bash
git clone https://github.com/hshintelligence/agent-scrape.gitcd agent-scrapenpm install
# Set secretsnpx wrangler secret put GROQ_API_KEYnpx wrangler secret put CDP_API_KEY_IDnpx wrangler secret put CDP_API_KEY_SECRET
# Local devnpx wrangler dev
# Deploynpx wrangler deploy

To use your own wallet as payTo, edit the PAY_TO constant in src/index.ts.

License

MIT — see LICENSE.

Contact

Built by HSH Intelligence, operated by Healing Sun Haven LLC (Wyoming, USA).

Pay-per-call web scraping for AI agents — no signup, no API keys, just USDC.

Using AgentScrape from ElizaOS

ElizaOS supports remote MCP servers natively via the @elizaos/plugin-mcp plugin. AgentScrape works out of the box — no custom integration needed.

Install the MCP plugin

bash
bun add @elizaos/plugin-mcp# ornpm install @elizaos/plugin-mcp

Configure your character to use AgentScrape

In your ElizaOS character JSON, add the plugin and point the MCP server config at AgentScrape's Streamable HTTP endpoint:

json
{  "name": "ResearchAgent",  "plugins": ["@elizaos/plugin-mcp"],  "settings": {    "mcp": {      "servers": {        "agentscrape": {          "type": "streamable-http",          "name": "AgentScrape",          "url": "https://agent-scrape.healingsunhaven.workers.dev/mcp",          "timeout": 60        }      }    }  }}

Available tools (auto-discovered by ElizaOS)

Once configured, the agent automatically discovers all six AgentScrape tools:

ToolCostPurpose
scrape_webpage$0.003 USDCMarkdown/HTML/text/JSON scrape
extract_structured_data$0.005 USDCAI extraction via Groq + Llama 4 Scout
screenshot_webpage$0.003 USDCPNG screenshot with viewport control
extract_metadata$0.002 USDCTitle, OG, Twitter, JSON-LD
create_browser_session$0.001 USDCStateful browser session
run_workflow$0.008 USDCMulti-step atomic workflow up to 20 steps

Free tier

The first 10 calls per wallet in the first 30 days are free. After that, calls are paid in USDC on Base mainnet (eip155:8453) via the x402 v2 payment protocol. To pay, supply the X-PAYMENT-RESPONSE header with an x402 payment receipt on retry — the standard x402 flow.

Example prompt

Once the plugin is wired up, your ElizaOS agent can use AgentScrape transparently:

User: "Scrape https://www.x402.org and give me a two-sentence summary."

ElizaOS routes the request through @elizaos/plugin-mcp, which calls AgentScrape's scrape_webpage tool, and the agent answers based on the live page content.

Resources

來源：README.md，提交 2636125

工具

6

scrape_webpage

Scrape Webpage

Scrape any webpage and return content as markdown, html, text, or json. Pay-per-call web scraping for AI agents.

輸入參數url必填formatwait_msviewportwait_for

extract_structured_data

Extract Structured Data

AI-powered structured data extraction from any webpage using natural language. Returns JSON matching your prompt or schema.

輸入參數url必填prompt必填schemawait_mswait_for

screenshot_webpage

Screenshot Webpage

Capture a PNG screenshot of any webpage. Supports desktop, mobile, and tablet viewports, plus full-page mode.

輸入參數url必填wait_msviewportwait_forfull_page

extract_metadata

Extract Page Metadata

Extract page metadata: title, description, Open Graph, Twitter cards, JSON-LD, canonical URL, and all meta tags.

輸入參數url必填

create_browser_session

Create Browser Session

Create a stateful browser session that persists cookies and localStorage across multiple scrape/workflow calls.

輸入參數ttl_seconds

run_workflow

Run Browser Workflow

Execute a multi-step browser workflow atomically: navigate, click, type, wait, scroll, screenshot, extract, evaluate. Up to 20 steps.

輸入參數steps必填viewportsession_idpersist_session

版本歷史

1
  1. v0.6.1最新Sep 16, 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 Scrape 整合

Scrape.ly

ly.scrape官方發布者

Scrape.ly 的 MCP 伺服器，讓助理在 Scrape.ly 商店中搜尋 Actor、執行它們，並讀取其資料集、鍵值儲存與日誌。

Streamable HTTP可網頁執行驗證方式
網頁搜尋與抓取需要憑證Nov 7, 2026

SEO Data MCP (SteadyScrape)

由 GitHub 帳號 steadyscrape 發布

關鍵字搜尋量、即時SERP排名、網域流量、排名關鍵字、反向連結、Google Trends、YouTube。AI 翻譯

Streamable HTTP可網頁執行驗證方式
網頁搜尋與抓取需要憑證Nov 3, 2026

x402-scrape-mcp

由 GitHub 帳號 6mickswagger9 發布

本機 MCP 伺服器，為 AI 代理提供 x402 付費 API，透過 CDP facilitator 以 Base 上的 USDC 結算。

STDIO僅桌面
網頁搜尋與抓取未驗證 · 伺服器未能啟動Oct 31, 2026

TrueScrape

由 GitHub 帳號 TrueScrape 發布

付費 API，讓助理取得 52 個平台的公開即時資料——社群帳號與貼文、商品列表、房產與廣告資料，並支援提及監控。

Streamable HTTP可網頁執行驗證方式
網頁搜尋與抓取200+ 個工具Nov 4, 2026

Jobs Data MCP (SteadyScrape)

由 GitHub 帳號 steadyscrape 發布

由 Apify 託管的遠端 MCP 伺服器，讓助理取得來自 14 個以上 ATS 招募網站及遠端職缺的即時徵才資訊。

Streamable HTTP可網頁執行驗證方式
網頁搜尋與抓取需要憑證Nov 6, 2026