Proofetch

io.github.Vektoris-AIv0.0.2更新於 Oct 9, 2026

Verified extraction: source-backed JSON from PDFs/URLs; honest null + signed receipt.

已驗證Streamable HTTP可網頁執行Files & StorageWeb Search & ScrapingData & Analytics

概覽

AI 產生的概覽

從 PDF 與網頁擷取結構化 JSON，每個欄位都有來源引文或誠實的 null，並附上簽章回執。

功能
Proofetch 是託管式擷取服務，以遠端 MCP 伺服器與 HTTP API 的形式提供。其工具 verified_extract 接收來源（pdf_url、pdf_base64、url、text 或 html）與 JSON Schema 格式的 targetSchema，回傳乾淨的 JSON，其中每個欄位都帶有 PASS 或 FAIL 的判定以及來自原始文件的來源引文。無法證實的欄位會回傳 null 而非猜測值，整個結果封裝在帶雜湊的 Ed25519 簽章回執中，可獨立驗證。
適用情境
當助理需要從發票、合約或網頁中擷取特定欄位，而且你需要知道哪些值確實有來源支持時使用。它適合錯誤或編造的值比缺漏值更糟、且逐欄位來源或可驗證回執很重要的情境。
執行需求
使用遠端端點，無需自行託管。需要以 Authorization Bearer 標頭傳送 API 金鑰，金鑰可透過在供應商落地頁購買預付餘額（最低 5 歐元）取得。需要能連線至託管端點的網路。
安裝前請注意
此服務需付費：依通過驗證的欄位計費（每個已驗證欄位 0.02 歐元），因此擷取呼叫會消耗預付餘額。API 金鑰是作為 Authorization Bearer 標頭傳送的憑證，應妥善保密。來源文件與 URL 會傳送至第三方託管服務處理，且該流程為專有並由遠端營運。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 Proofetch，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。

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "proofetch": {
      "type": "http",
      "url": "https://proofetch-vektoris-ai-workflows.vercel.app/mcp"
    }
  }
}

README

proofetch — verified extraction with proof per field

Structured data from PDFs and web pages, where every value is backed by an exact source quote — or honestly left null. Never guessed. Each result ships with a cryptographically signed receipt you can verify yourself.

proofetch is a hosted MCP server and HTTP API built for developers and AI agents. You pay only for fields that pass the proof — €0.02 per verified field. Empty or unprovable fields cost nothing.

  • 🔗 Landing page & pricing: https://vektoris.de/proofetch
  • 🤖 MCP endpoint (remote): https://proofetch-vektoris-ai-workflows.vercel.app/mcp
  • 🧾 Listed in the official MCP registry: io.github.Vektoris-AI/proofetch

This repository is the public connector & documentation for the hosted proofetch service. The extraction pipeline itself is operated as a service — you don't self-host it, you call it with an API key.

Why proofetch

Normal extraction fills every field — even when the value is invented. You then have to re-check every number by hand. proofetch is different:

  • Never guesses. Unsure? The field is null + flagged, instead of a fabricated value.
  • Provenance per field. Every value comes with the exact source quote, verifiable against the original document.
  • Signed receipt. The result is sealed in a signed receipt (Ed25519) — you, or an escrow, verify it without having to trust us.
  • Pay for truth only. Billing is per field that passes the proof (verdict: PASS).

Get an API key

Buy prepaid balance (from €5) on the landing page — you get your key instantly after payment:

👉 https://vektoris.de/proofetch#preise

Quickstart — HTTP API

bash
curl -X POST https://proofetch-vektoris-ai-workflows.vercel.app/extract \  -H "Authorization: Bearer YOUR_API_KEY" \  -H "Content-Type: application/json" \  -d '{    "source": { "type": "pdf_url", "value": "https://example.com/invoice.pdf" },    "targetSchema": {      "type": "object",      "properties": {        "invoice_total": { "type": "string" },        "due_date":      { "type": "string" },        "invoice_number":{ "type": "string" }      },      "additionalProperties": false    }  }'

Source types (source.type): pdf_url, pdf_base64, url, text, html. targetSchema is a standard JSON Schema (type: "object", properties, additionalProperties: false).

What you get back

jsonc
{  "data": {    "invoice_total": "1,190.00 EUR",    "due_date": "2024-09-30",    "invoice_number": null          // honest null — not found in source  },  "fields": [    { "path": "invoice_total", "value": "1,190.00 EUR", "verdict": "PASS",      "provenance": { "quote": "1,190.00 EUR", "start": 43, "end": 55 } },    { "path": "invoice_number", "value": null, "verdict": "FAIL", "provenance": null }  ],  "verdict": "PARTIAL",  "receipt": { "v": 1, "jobId": "pf_…", "outputHash": "sha256:…" },  "billing": { "charged": true, "amountCents": 4 }}
  • data — clean JSON, each field proven or honestly null
  • fields[] — per field: source quote (provenance) + verdict (PASS/FAIL)
  • verdict — overall: PASS, PARTIAL, or FAIL
  • receipt — signed proof with hashes, self-verifiable

Check your balance any time: GET /billing/balance with the same key.

Use as an MCP server (remote)

proofetch speaks the Model Context Protocol over Streamable HTTP. Point any MCP-capable client at the remote endpoint with your key as a bearer token.

Endpoint: https://proofetch-vektoris-ai-workflows.vercel.app/mcp Auth: Authorization: Bearer YOUR_API_KEY Tool: verified_extract — same contract as the HTTP API above.

Example client config (generic MCP streamable-http remote):

json
{  "mcpServers": {    "proofetch": {      "type": "streamable-http",      "url": "https://proofetch-vektoris-ai-workflows.vercel.app/mcp",      "headers": { "Authorization": "Bearer YOUR_API_KEY" }    }  }}

Open standards

proofetch is built for autonomous agents: discoverable via the A2A card and MCP, payable per call via x402 / AP2, and every receipt is independently verifiable.

Links

© Vektoris. The proofetch service and pipeline are proprietary; this connector/documentation is provided under the MIT License (see LICENSE).

來源：README.md，提交 52ce974

工具

0
工具後設資料尚未被收錄。

版本歷史

1
  1. v0.0.2最新Sep 16, 2026