Agent Data Tools: Documents to text

io.github.Nutshell2000由 GitHub 账号 Nutshell2000 发布v1.179.0更新于 Nov 14, 2026

5 tools: PDF, Word, Excel, PowerPoint and web pages to Markdown; invoice fields as JSON.

已验证Streamable HTTP可网页运行网页搜索与抓取文件与存储

概览

AI 生成的概览

按次付费的远程 MCP 服务器,把 PDF、Office 文档、扫描件和网页转成干净的文本、Markdown 或 JSON 字段。

可以做什么
  • 抓取公开网页并输出干净的文本或 Markdown,另有浏览器渲染版本处理 JavaScript 页面
  • 逐页读取公开 PDF 的文本,并给出标题、作者和页数
  • 把 Word、Excel、PowerPoint、PDF 以及 OCR 后的扫描件或图片转成文本或 Markdown
  • 从文档中按指定字段名提取信息并以 JSON 返回,例如发票号、金额、到期日
适用场景
适合助手需要读取自己无法打开的文件或网页时:从 PDF 和 Office 文件中取出文本作为 RAG 上下文、把网页转成 Markdown,或把发票等文档转成结构化 JSON 字段。
运行要求
远程 Streamable HTTP 端点,无需本地运行时或注册。每个调用方有 25 次免费成功调用,之后需付费:用 Authorization Bearer 头传预付卡密钥,或使用支持 x402/MPP 的客户端以 USDC 支付。仅支持公开 URL 或上传文件,不支持需登录的页面。
安装前请注意
免费额度用完后每次成功调用收费(0.002 至 0.03 美元),Authorization 密钥是付费凭证需妥善保管。上传的文档和抓取的网址会发送给第三方服务,请勿处理机密文件。仅限公开页面和文件,且有大小和字符数上限。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Agent Data Tools: Documents to text,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.Nutshell2000%2Fagent-data-tools-documents/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "agent-data-tools-documents-to-text": {
      "type": "http",
      "url": "https://agent-data-tools.revmesh2074.workers.dev/mcp/documents"
    }
  }
}

README

Agent Data Tools

[Add to Cursor] [Install in VS Code] [Install in VS Code Insiders]

187 pay-per-call tools for AI agents over MCP and plain HTTP: web search and news, payee screening and OFAC sanctions checks, crypto token prices, wallet balances and token transfers, web page to Markdown, PDF to text, email verification, company and domain lookups, weather, SEC filings, World Bank GDP, inflation, population and unemployment, LLM chat and embeddings. In full: blockchain reads on 5 EVM chains plus Bitcoin, token prices and risk, OFAC sanctions and counterparty checks, company and domain data, web pages and PDFs as text, weather, time, countries, and everyday utilities (JSON, CSV, hashes, dates, validators). Most tools cost $0.01 per successful call.

APIs for developers, by monthly plan

One API key, from $9 a month or pay as you go, paid by card. Each page has the request in curl, Python and Node.js and a free try:

Quick start

  1. Connect any MCP client to https://agent-data-tools.revmesh2074.workers.dev/mcp/compact (Streamable HTTP, no sign-up): three tools that find, describe and run any tool in the catalog, so your assistant's context stays small. The buttons above connect it too. Claude Code: claude mcp add --transport http agent-data-tools https://agent-data-tools.revmesh2074.workers.dev/mcp/compact. Every tool as its own definition: https://agent-data-tools.revmesh2074.workers.dev/mcp. From a terminal, for Claude Code, Claude Desktop, Cursor or VS Code: npx -y agent-data-tools setup. Claude.ai and ChatGPT: add a custom connector with https://agent-data-tools.revmesh2074.workers.dev/mcp/account. Or connect one section: /mcp/prices, /mcp/chain, /mcp/web, /mcp/compliance, /mcp/ai, /mcp/everyday, /mcp/utilities.
  2. Try free: every caller gets 25 free successful calls over MCP, used automatically, or 3 over HTTP with trial=1 (PDF reading is not free over HTTP). Failed calls never count. AI, rendering and search tools are not included.
  3. Then pay either way:
    • Card: buy prepaid credits at /credits ($5 = 5,000 credits, $20 = 22,000, $100 = 125,000; 1 credit = $0.001) and send the key as Authorization: Bearer adt_… (or X-API-Key) on HTTP calls or in your MCP client's connection headers.
    • Crypto: any x402 client pays per call in USDC on Base, Polygon, Arbitrum or Solana; MPP clients pay on Tempo or Base.

Failed calls are never charged. Free previews of every tool's output: add sample=1. Machine-readable catalog: /openapi.json, /llms.txt. Support: [email protected].

Details and buyer clients

Operated by SpellzCode. Check a Base transaction or wallet through a capped, account-free client: one transaction-status result costs $0.003, a wallet snapshot $0.005. Inspect a free quote before an explicitly authorized purchase. Optional gas and allowance checks cost $0.002 each; the client never buys them automatically. Uses an existing funded Base x402 wallet and never sends the transaction being checked.

For static documentation reading, get clean text, Markdown, code and tables for $0.002 USDC on Base. No account or subscription. Static HTML, Markdown and plain text only; no JavaScript rendering, authenticated pages or paywall access.

The reader client requires Node 20.19+ and no npm installation or TypeScript. An unsigned quote needs no key and never pays. An explicit authorized reader purchase needs an existing funded Base wallet and a per-call cap of at most $0.002. The local MCP adapter starts in quotes-only mode; deliberately enabled buying shares a process budget across reader, sitemap and Bitcoin-data calls. Keys stay on the buyer's device. The separate chain client has the operation-specific caps above and grants no session or monthly spending allowance.

Repository fallback downloads for 1.7.6: CLI executable and its checksum, or local MCP adapter and its checksum. Save the executable and matching checksum in one folder, then follow the verification instructions before running it. Release notes and all six files. These match the website's published bytes; checksums are not independent security signatures. Older 1.7.0 downloads and 1.5.1 downloads remain available unchanged.

Need help with an existing attempt? Report a purchase problem using the request ID, approximate UTC time and reported payment status. Stop after an uncertain signed outcome; do not make a new purchase to reproduce it. For missing text, code or tables, report an output problem with a short public or authored example. Reports are public: never include keys, signatures, raw payment headers or private content.

The reader also supports an exact HTML section ID and retains labelled source image descriptions; it does not interpret image pixels. Source-declared authors and publication/update dates now include provenance and unverified/ambiguity warnings when available. Check scope, output limits and warnings before claiming coverage. Full buying instructions.

Cached examples are actual build-time results on owned HTML and are labelled separately from illustrative snippets. Local tests do not establish funded production delivery or customer demand.

A Cloudflare Worker that sells more than 600 tools to AI agents. Each call is paid in USDC through the x402 protocol (Base, Polygon, Arbitrum or Solana), through MPP (Tempo), or with prepaid card credits, and the money goes straight to the address in PAY_TO.

EndpointReturnsPrice
GET /sitemap?url=&contains=&maxUrls=&offset=XML sitemap URLs, source-reported dates and pagination (bounded, no crawling)$0.002
GET /extract?url=Web page as clean text, headings and links$0.002
GET /domain?name=DNS, registration age, SPF/DMARC$0.005
GET /domain-report?name=Domain trust verdict and email-auth flags$0.01
GET /token?address=Base ERC-20 facts and risk flags$0.01
GET /token-report?address=Base token risks and live DEX market data$0.01
GET /wallet?address=Base wallet balances and account nonce$0.005
GET /summarize?url=Short summary and key points of a page$0.01
GET /screenshot?url=JPEG of the rendered page (base64)$0.01
GET /image?prompt=Generated 1024x1024 image (base64)$0.01
GET /transcribe?url=Text of an audio file (max 5 MB)$0.01
GET /describe?url=Description of an image, or an answer about it$0.01
GET /btc-feesBitcoin fee rates and mempool status$0.002
GET /btc-address?address=Bitcoin address balance and activity$0.01
GET /ln-invoice?invoice=Decoded Lightning invoice
GET /package-audit?packages=Audit up to 10 dependencies; supports exact-version advisories and fixes$0.01
GET /wallet-audit?address=Base token holdings and nonce-based heuristic risk indicators$0.01
GET /contract-risk?address=Limited bytecode, EIP-1967 slot and zero-supply heuristics$0.01
GET /trade-insight?token=&wallet=Separate token and wallet risk indicators; no trade recommendation$0.01
GET /liquidity-pool-audit?address=Verified Base DEX pair details and reported market data$0.01
GET /mev-simulatorBase gas-cost estimates using fixed typical gas units; not a quote$0.01
GET /package?ecosystem=&name=&version=Latest or pinned package metadata and bounded dependency-risk signals (not a safety guarantee)
GET /email?address=Email validation without sending mail
GET /papers?query=Academic paper search$0.01
GET /fx?from=&to=Currency exchange rates and conversion
GET /btc-priceBitcoin price in major currencies
GET /base-gasBase network gas price and fee estimate$0.002
GET /base-tx?hash=Base transaction status and details$0.003
GET /base-allowance?token=&owner=&spender=ERC-20 approval check on Base$0.002
GET /ens?name=ENS name/address resolution
GET /weather?lat=&lon=Current weather and hourly forecast
GET /time?zone=Current time and time zone conversion

Free calls are on: 25 successful calls per caller over MCP, 3 over HTTP (not /pdf/text), and 2,000 a day in total (src/lib/free-calls.ts), automatic over MCP and with trial=1 over HTTP. Without trial=1, an unpaid HTTP request gets the normal 402 challenge. The table above lists a few core tools; the full, current catalog is at /#tools and in /openapi.json.

The homepage and llms.txt include a copy-pastable Node.js/TypeScript x402 client using @x402/fetch, @x402/evm and viem. It reads the private key from X402_PRIVATE_KEY (never hard-code or commit it), pins Base, the service's published PAY_TO recipient and Base USDC, and caps each payment at $0.01, the ordinary per-call price, so it can pay for every tool priced at $0.01 or less. It does not add trial=1. /package-audit remains $0.20 per call and is intentionally refused by this lower-spend example. The OpenAPI x-guidance carries the same client, price and trial notes. Tool-specific guides (/guides/web-reading, /guides/bitcoin-fees, /guides/dependency-audit) pin that one tool's exact price instead.

A staging copy of the service runs on the Base Sepolia testnet (wrangler.staging.jsonc, keyless facilitator, faucet USDC) so every release can be paid for real over HTTP v2, HTTP v1 and MCP with scripts/pay-test.mjs before production deploys; see CLAUDE.md. Testnet settlements are never income.

/package keeps its latest-version behavior when version is omitted. Pass ecosystem, name and an optional exact version as separate inputs (for example, ecosystem=npm&name=lodash&version=4.17.20) to return pinned-version metadata and deps.dev advisories. Version ranges and tags are rejected. For exact-version OSV affected ranges and dated FIRST EPSS CVE estimates, use /package-audit.

/package-audit accepts up to 10 comma-separated ecosystem:name items, optionally with an exact version, such as npm:[email protected] or npm:@types/[email protected]. Unversioned items retain latest-version checks. To bound outbound fetches, it does not query npm weekly downloads; weeklyDownloadsStatus is not-fetched-subrequest-budget. Unversioned items fetch at most one deps.dev advisory detail and report advisoryDetailsOmitted; failed detail fetches are counted in advisoryDetailsUnavailable and make advisoryDetailsStatus partial. Pinned items skip duplicate deps.dev details but retain all advisory IDs/counts. Exact versions use OSV batch IDs and up to 15 detailed records per request; per-advisory detailStatus, detailsNotFetched and omittedAdvisories identify incomplete coverage. Returned CVE aliases are enriched with dated FIRST EPSS estimates in one deduplicated request (up to 50 CVEs); epssLookup reports source status, retrieval time, omitted CVEs and uninspected OSV details. EPSS estimates 30-day exploitation probability, not package safety; missing scores are unknown, not zero. Neither ok nor an empty advisory result means a package is safe.

/airdrop-scanner, /liquidity-return-analyzer, /token-holder-analysis, /pair-analyzer and /wallet-history are deferred because this service has no verified live data sources for their advertised analyses. They are not listed as paid tools and return HTTP 503 without a payment challenge or charge.

The AI endpoints use Workers AI and /screenshot uses Browser Rendering. Account allowances and actual operating costs must be checked separately; they are not established by the API price. Upstream capacity or rate-limit failures return errors without settling a charge.

Free routes include the owned /examples/extract/ results, versioned /clients/reader/ downloads, /sitemap.xml, / (a web page for browsers, JSON for programs), /openapi.json, /llms.txt, /robots.txt, /health, /.well-known/x402, /.well-known/mcp/server-card.json, /.well-known/api-catalog.

/metrics reports daily request counts by path. unsettled counts successful HTTP 2xx responses on paid paths without a verified successful PAYMENT-RESPONSE; only calls with a valid settlement response increment paid. revenueUsdMicros is gross USD in millionths of a dollar, and trackedPaidCalls counts settled x402 HTTP API, MCP tool and /buy/<slug> calls included in that total. MCP totals are recorded only when the payment wrapper returns a successful structured settlement result; x402 product purchases use the exact product catalog price. Revenue fields are prospective and have no historical backfill. paid counts from before settlement validation are not proof of payment and must not be used as revenue. Challenges, trials, marketplace requests, unsettled responses, and direct storefront-transfer amounts are excluded.

Store

New digital-product sales are paused at /store and /buy/* (HTTP 503, no payment challenge). Existing orders and downloads are preserved. The retained implementation works as follows: A buyer gets an exact USDC amount and the address in PAY_TO; the order page watches the chain and unlocks the download when a matching transfer arrives. Agents can buy the same products at GET /buy/<slug> through the normal paywall.

  • Products are listed in src/store/catalog.ts.
  • Orders and the sales log live in the TRIALS key-value store (order: and paid: keys). List sales with npx wrangler kv key list --binding TRIALS --prefix paid: --remote.
  • The DeFi Trader Intelligence Bundle is delisted because the service has no API-key authentication, 250-call quota or /account endpoint. Legacy order receipts remain supported, but actual download fulfillment is unverified. Product assets under the gitignored public/_files/ directory are intentionally absent from Git and must be privately provisioned before deploying a fresh checkout; missing assets make paid downloads return 502. The existing production worktree retains these assets. Never publish them in source control.

MCP

All 179 catalog tools are available as an MCP server at /mcp (streamable HTTP, stateless), in smaller section servers at /mcp/<section>, and behind three tools at /mcp/compact (find_tool, tool_details, call_tool: the whole catalog without hundreds of tool definitions in the assistant's context; same prices, same free calls). Listing tools is free. Every MCP caller gets 25 free successful calls; after that a call uses card credits (an API key from /credits in the connection headers) or returns an x402 payment requirement that an MCP client with x402 support pays and retries. Signed-in connectors at /mcp/account never see crypto payment requests. A tool call that fails is not charged.

Each tool is defined once in src/endpoints.ts (name, price, parameters) and implemented once in src/tools.ts; the HTTP route, MCP tool, OpenAPI document and web page are generated from that catalog.

Where it is listed

  • Official MCP Registry: listed as io.github.Nutshell2000/agent-data-tools. To publish a new version, raise version in server.json and push a matching v* tag; the workflow in .github/workflows/publish-mcp.yml does the rest on GitHub's servers.
  • 402 Index (https://402index.io): existing reader entries and catalog metadata were refreshed on 15 October 2026. Third-party caches can remain stale; the live catalog and quotes take precedence.
  • x402scan: existing listings cover the catalog. Reconcile current prices after each release; no new duplicate registration is needed.
  • Coinbase Bazaar: Bitcoin fees and package metadata have indexed entries. Other routes require a genuine settled payment; a readiness pass alone does not index them.

An unpaid request gets HTTP 402 with a PAYMENT-REQUIRED header (x402) and WWW-Authenticate: Payment challenges (MPP: evm with USDC on Base, tempo with USDC.e on Tempo; settled only after the tool succeeded). Calls that fail (4xx/5xx) are not charged.

OpenAI-compatible: POST /v1/chat/completions, POST /v1/responses and POST /v1/embeddings are fronts for the paid /ai/chat, /ai/chat-large and /ai/embed tools (same prices, same payment paths); GET /v1/models is free. Point an OpenAI SDK at /v1 with an x402 or MPP fetch wrapper, or use a card-credit key as its apiKey. Clients that can only GET send the same inputs in the query (/v1/chat/completions?prompt=…, /v1/responses?input=…, /v1/embeddings?input=…).

Current revenue operations

New seller-kit sales are paused; existing paid orders keep download access. Free calls are on (see above) and are not revenue. /metrics reports request counters; /revenue reports deduplicated, facilitator-reported settlements observed since the ledger was deployed. Its amounts are integer native Base USDC atomic units (6 decimals), and known self-payments are excluded from gross. Receipts preserve transaction, payer, authorization nonce, endpoint and amount evidence privately in the existing Durable Object. Signatures and request content are not stored. Public responses expose aggregate totals only. A storage error is logged without turning an already-settled customer response into another payment request.

These totals remain pending independent chain reconciliation. Fees, expenses and net income are not inferred. Historical paid counters cannot be backfilled using today's prices. The two indexed historical receipts were 2,000 atomic units each and remain separate from the current settlement ledger. Tests and self-payments are never customer income.

The live catalog is authoritative for current tools, prices and trial availability.

Configuration

All settings are in wrangler.jsonc:

  • PAY_TO: the address that receives payments.
  • NETWORK: eip155:8453 for Base mainnet.
  • FACILITATOR_URL: the service that verifies and broadcasts payments.
  • PAYWALL: set to off only for local testing.
  • TRIALS_ENABLED: trials are disabled unless set exactly to on; set to off (or unset it) to disable them.

Commands

bash
npx wrangler dev

Read the current Worker version with npx wrangler deployments list, then deploy from a clean checkout containing current origin/main and successful CI for its exact commit:

sh
npm run deploy:production -- --expected-version <live-worker-version-uuid>

Add --check-only for preflight without deployment. A stale checkout replaced the buying guides and ledger on 17 October; the guard rejects stale or uncommitted source and an unexpected production baseline. It cannot lock out another operator bypassing this process. Coordinate production deployments; preserve concurrent work in other checkouts.

bash
npx wrangler tail

wrangler tail streams live requests, which is the quickest way to see whether anyone is calling the service.

On Windows, wrangler dev can fail with SQLITE_CANTOPEN when the project path is long. Add --persist-to with a short folder path to fix it.

To try /screenshot without the paywall, use npx wrangler dev --remote --var PAYWALL:off. It runs on Cloudflare, so no browser is downloaded to this PC.

Listing in the Coinbase Bazaar

This deployment already uses the authenticated Coinbase CDP facilitator. For a separate installation, configure your own account:

  1. Create a free API key at https://portal.cdp.coinbase.com.

  2. Store it as Worker secrets (you'll be prompted for each value):

    bash
    npx wrangler secret put CDP_API_KEY_ID
    bash
    npx wrangler secret put CDP_API_KEY_SECRET
  3. In wrangler.jsonc, set FACILITATOR_URL to https://api.cdp.coinbase.com/platform/v2/x402 and deploy.

The Bazaar indexes an endpoint after a successful settled payment with discovery metadata. Never manufacture self-purchases to seed it. npm run check:live runs the free, read-only Coinbase validator against every documented tool example. It does not sign, settle or create an index entry. The historical scripts/index-listings.mjs --pay route is disabled. HTTP and MCP funded production delivery remain separate evidence requirements.

Checking that it works

Live URL: https://agent-data-tools.revmesh2074.workers.dev

Coinbase runs a free validator. A pass shows "valid":true and "simulation":{"outcome":"accepted"}:

bash
curl -X POST https://api.cdp.coinbase.com/platform/v2/x402/validate -H "Content-Type: application/json" -d "{\"resource\":\"https://agent-data-tools.revmesh2074.workers.dev/domain?name=cloudflare.com\",\"method\":\"GET\"}"

Incoming payments show up as USDC transfers to PAY_TO on https://basescan.org.

Limits

  • /extract reads at most 2 MB of a page and waits at most 8 seconds for it.
  • /token and /wallet use the public Base RPC, which is rate limited. Set BASE_RPC_URL to a private endpoint if volume grows.
  • Token capability flags are bytecode heuristics, and for a proxy they describe the proxy, not its implementation.

Agent-client compatibility evidence

On 15 October 2026, @agentcash/discovery 1.7.5 discovered all 31 live routes from our OpenAPI, recognized /extract as paid x402 at 0.002 USD, retrieved the standalone-client guidance and read the full reader schema. This was a wallet-free metadata check, not funded delivery or a customer purchase. AgentCash CLI 0.17.1's check command loads an existing wallet; its discovery library can inspect schemas without one. Our standalone quote needs no wallet or key.

Reproduce the read-only check with an independently installed discovery library:

sh
node scripts/check-agent-discovery.mjs /path/to/@agentcash/discovery/dist/index.js /private/path/evidence.json

The script does not install packages, initialize wallets, sign payments or execute paid extraction. It checks the current production catalog, price, guidance and structured output schema, plus what x402scan sees: a live 402 probe, the readable 402 body, the HTML homepage with og:image, /og.png and /status. Set CHECK_ORIGIN to check another deployment. Funded HTTP, CLI and MCP delivery require separate evidence.

Use a local MCP agent without writing a client

The bundled local MCP adapter runs with Node 20.19+ and no npm installation. Quotes-only by default. Explicit paid mode enables one-call purchases capped at $0.002 each and a process session allowance; unknown signed outcomes lock further buys. Source and setup. Preserve host tool approvals. The remote MCP connection alone does not pay automatically.

Read the relevant section of a long document

Optional section selects an exact visible heading/container ID within primary HTML. Use --section with the CLI or section with MCP. Missing/ambiguous selections fail uncharged; returned scope is explicit. Actual cached late-section example and scope instructions. Price remains $0.002. URL fragments alone do not change scope.

Portable buyer skill

The single-file reader skill gives a compatible agent the quote, authorized purchase, section-reading and failure-reconciliation workflow. Owned download and installation context. Place it in a folder named spellzcode-x402-reader in your host's configured skills location using that host's procedure. Inspect before installing; no automatic installation, wallet creation or spending authorization is supplied. Host activation and funded delivery are not established by publication.

Secondary entry: Bitcoin fee data

The same downloaded client now supports node reader-client.mjs quote-btc-fees and one authorized node reader-client.mjs buy-btc-fees --max-usdc 0.002. No URL, npm setup or Bitcoin key. Local MCP exposes btc_fees_quote/btc_fees_buy under the shared session allowance. This buys data for Base USDC and never sends Bitcoin. Setup and limitations. The static reader remains the primary workflow.

Earlier reader releases

Version 1.2.0 adds the bounded /sitemap tool and capped quote-sitemap/buy-sitemap commands. Version 1.1.2 adds a local command builder with POSIX/PowerShell quoting and a three-page research recipe, while preserving immutable 1.1.1 downloads. Reader 1.1.1 separates metadata limits from primary table text, preserves literal code whitespace and nested lists, and closes clipped code fences within the output limit. The bundled Node client supports an unsigned quote and one explicitly capped buy. Build it and the cached samples with npm run build:reader. Downloaded clients need Node 20.19+ and no project dependencies. Checksums and public source accompany the download; keys stay in the buyer's process environment.

Remote MCP advertises successful-result JSON schemas for extract, sitemap and btc-fees, matching OpenAPI. Results are checked before settlement. Quotes and failures return isError: true with JSON in content[0].text; only successful results have structuredContent. This keeps free quotes usable by schema-validating clients.

Endpoint families

MCP catalog 1.8.1 adds 113 active family tools to the 32 existing tools: blockchain reads across five networks, local data/text utilities, public-data lookups and crypto-random utilities. Active family calls cost $0.002. The live OpenAPI and catalog list the current routes. Reader downloads stay at 1.7.6 and the separate chain client stays at 1.0.0.

Four implemented family tools are paused and excluded from paid discovery: /regex/test until execution can be bounded, and /air-quality, /uv, /geocode until a source permitted for this commercial service is configured. Direct requests return uncharged 503 without calling the upstream. No subscription or new spending was added. JSON repair preserves already-valid JSON and quoted text.

来源:README.md,提交 1fa2edd

工具

5

extract

Scrape website page to Markdown: page URL into clean Markdown or text for LLMs

Fetch any public web page URL and get clean, LLM-ready text or Markdown: main content, code blocks, tables, links and headings, for RAG and agent context. Static HTML, Markdown or text; optional section ID. Default 20,000 output characters, max 100,000; input 2 MiB. No JavaScript rendering (see /extract/rendered), logins or paywalls. Empty pages, bad sections, browser challenges and errors are not charged. Use to read docs or articles. Example: url=https://example.com. Price $0.002/call via x402 USDC or a card key (/credits), first 25 free; failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.

输入参数url必填formatsectionmaxChars

extract_rendered

Web page to Markdown, rendered: a JavaScript-rendered page as text or Markdown

Same output as /extract, but the page is first rendered in a real browser, so single-page apps and pages that build their content with JavaScript come back with their text instead of an empty shell. Waits for the network to settle (up to 15 seconds). Public pages only; no login. Use when /extract returns EXTRACTION_EMPTY or too little text for a page you can read in a browser. Example: url=https://example.com/. Price $0.01/call via x402 USDC or a card key (/credits); failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.

输入参数url必填formatsectionmaxChars

pdf_text

PDF to text, PDF text extraction: extract the text of a PDF document from a URL, page by page

Extract the text of a public PDF (up to 10 MB) from its URL, page by page, with title, author and page count. Choose the first page, how many pages (up to 200) and maxChars. arXiv abstract links work. If a site blocks our fetch, the Internet Archive's dated copy is read and the answer says so. Scanned, encrypted and non-PDF files fail uncharged. Use to read papers, invoices, reports and manuals into an agent. Example: url=https://bitcoin.org/bitcoin.pdf. Price $0.005/call via x402 USDC or a card key (/credits), first 25 free; failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.

输入参数url必填pagesmaxCharsfirstPage

document_text

Document to text: Word, Excel, PowerPoint, PDF

Convert a document to text or Markdown: Word (.docx) with headings, lists and tables, Excel (.xlsx) sheet by sheet, PowerPoint (.pptx) slide by slide, a PDF's text, and scanned PDFs or images (PNG, JPEG, WebP) by OCR. Give a public file's URL, or send the file in a POST body (10 MB; images 3 MB). Old .doc/.xls/.ppt and password-protected files fail uncharged. Use to read attachments, scans and spreadsheets into an agent, RAG or a workflow. Example: url=https://calibre-ebook.com/downloads/demos/demo.docx. Price $0.005/call via x402 USDC or a card key (/credits), first 25 free; failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.

输入参数url必填formatmaxChars

document_extract

Extract data from a document: invoice, receipt, contract fields as JSON

Extract named fields from a document as JSON: give a PDF, Word, Excel or PowerPoint file, a scan or a photo (PNG, JPEG, WebP) by URL or in a POST body (up to 10 MB; images 3 MB) and the field names you want, such as invoice_number, total, due_date. An AI model reads the file, scans by OCR, and returns one JSON object, with null where a value is not stated. Use to turn invoices, receipts, orders and contracts into rows. Price $0.03/call via x402 USDC or a card key (/credits), first 25 free; failed calls free. Reads or computes only and changes nothing; each call is charged, so repeats are not free.

输入参数url必填fields必填

版本历史

2
  1. v1.179.0最新Nov 14, 2026
  2. v1.177.0Nov 14, 2026

评分

已满足适用检查项的 100%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具必需15 分
  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分