Invoice Parser MCP

发票解析器MCP

从您的AI代理将发票、收据和财务文档解析为结构化JSON。

](https://smithery.ai/server/@knportal/invoice-parser-mcp)  ](CHANGELOG.md) 

上市时间： 格拉玛 · 史密瑟里 · mcp.so

为 模型上下文协议由Claude Vision提供技术支持。

它的作用

从PDF发票、扫描的收据和图像文件中提取结构化数据。无模板，无OCR配置——Claude Vision读取文档并返回干净的JSON。

______________________________________________________________________

工具

______________________________________________________________________

工具参考

parse_invoice

完整发票解析。从供应商发票中提取每个结构化字段。

参数

\*要么 api_key 或 payment_proof 必须提供。

示例输入

{
  "file_path": "/Users/me/documents/invoice_acme_jan2025.pdf",
  "api_key": "ip_free_abc123"
}

输出示例

{
  "ok": true,
  "document_type": "invoice",
  "vendor": {
    "name": "Acme Corp",
    "address": "123 Industrial Way, Austin TX 78701",
    "email": "billing@acme.com",
    "phone": "512-555-0100",
    "tax_id": "12-3456789"
  },
  "bill_to": {
    "name": "Jane Smith",
    "address": "456 Oak Ave, Boston MA 02101",
    "email": "jane@example.com"
  },
  "invoice_number": "INV-2025-0142",
  "invoice_date": "2025-01-15",
  "due_date": "2025-02-15",
  "payment_terms": "Net 30",
  "currency": "USD",
  "line_items": [
    {
      "description": "Software consulting — January",
      "quantity": 40,
      "unit_price": 175.0,
      "total": 7000.0,
      "tax_rate": 0.0
    }
  ],
  "subtotal": 7000.0,
  "discount": 0.0,
  "tax_amount": 560.0,
  "shipping": 0.0,
  "total": 7560.0,
  "amount_due": 7560.0,
  "notes": "Wire transfer preferred. See banking details on page 2.",
  "po_number": "PO-98765"
}

______________________________________________________________________

parse_receipt

解析零售或费用收据。专为销售点收据、餐厅账单和费用报销文件而设计。

参数

示例输入

{
  "file_path": "/Users/me/receipts/coffee_shop_march1.jpg",
  "api_key": "ip_free_abc123"
}

输出示例

{
  "ok": true,
  "document_type": "receipt",
  "merchant": {
    "name": "Blue Bottle Coffee",
    "address": "300 Webster St, Oakland CA 94609",
    "phone": "510-555-0200",
    "website": "bluebottlecoffee.com"
  },
  "date": "2025-03-01",
  "time": "09:14",
  "receipt_number": "5541",
  "cashier": "Maria",
  "items": [
    { "name": "Latte (large)", "quantity": 1, "unit_price": 6.50, "total": 6.50, "sku": "", "category": "beverage" },
    { "name": "Croissant", "quantity": 1, "unit_price": 4.00, "total": 4.00, "sku": "", "category": "pastry" }
  ],
  "subtotal": 10.50,
  "discounts": 0.0,
  "tax": 0.84,
  "tip": 2.00,
  "total": 13.34,
  "currency": "USD",
  "payment_method": "Visa",
  "card_last_four": "4242",
  "transaction_id": "TXN-88821",
  "loyalty_points": null,
  "notes": ""
}

______________________________________________________________________

extract_line_items

轻量级提取，仅返回逐项列表。比…更快、更便宜 parse_invoice 当你只需要行项目时。

参数

输出示例

{
  "ok": true,
  "line_items": [
    { "description": "Widget A (x10)", "quantity": 10, "unit_price": 12.00, "total": 120.00 },
    { "description": "Widget B (x5)",  "quantity": 5,  "unit_price": 24.00, "total": 120.00 }
  ],
  "item_count": 2
}

______________________________________________________________________

extract_totals

仅提取财务摘要（小计、税款、总计、到期日），不解析行项目或供应商详细信息。

参数

输出示例

{
  "ok": true,
  "currency": "USD",
  "subtotal": 240.00,
  "discount": 0.0,
  "tax_amount": 19.20,
  "tax_rate": 8.0,
  "shipping": 0.0,
  "tip": 0.0,
  "total": 259.20,
  "amount_due": 259.20,
  "invoice_date": "2025-01-15",
  "due_date": "2025-02-15"
}

______________________________________________________________________

validate_invoice

数学验证工具。验证行项目总计是否相等 quantity × unit_price，小计与行项目之和匹配，税务计算一致，最终总额对账。允许±0.02美元的舍入公差。

参数

输出示例——有效发票

{
  "ok": true,
  "valid": true,
  "issues": [],
  "summary": {
    "line_items_checked": 3,
    "subtotal": 450.00,
    "tax": 36.00,
    "total": 486.00,
    "currency": "USD"
  }
}

输出示例——无效发票

{
  "ok": true,
  "valid": false,
  "issues": [
    {
      "field": "line_item_2_total",
      "expected": 120.00,
      "found": 100.00,
      "description": "quantity (10) × unit_price (12.00) = 120.00, but stated total is 100.00"
    }
  ],
  "summary": {
    "line_items_checked": 3,
    "subtotal": 340.00,
    "tax": 27.20,
    "total": 367.20,
    "currency": "USD"
  }
}

______________________________________________________________________

export_to_csv

批量解析多达20张发票或收据，并导出摘要CSV。每行包含：文件名、文档类型、供应商/商家、日期、编号、小计、税款、总计、货币、到期日、付款方式。

参数

示例输入

{
  "file_paths": [
    "/Users/me/invoices/jan2025.pdf",
    "/Users/me/invoices/feb2025.pdf",
    "/Users/me/receipts/expense_march.jpg"
  ],
  "output_path": "/Users/me/exports/q1_summary.csv",
  "api_key": "ip_free_abc123"
}

示例输出（JSON响应）

{
  "ok": true,
  "output_path": "/Users/me/exports/q1_summary.csv",
  "rows_written": 3,
  "errors": []
}

CSV列： filename, document_type, vendor_merchant, date, number, subtotal, tax, total, currency, due_date, payment_method

______________________________________________________________________

支持格式

• PDF（发票、扫描文件）
• PNG、JPG、WEBP（收据照片、截图）

______________________________________________________________________

认证

免费等级： 使用API密钥每月20次解析（在pleitudo.ai获取一个）

按次付费（x402）： 不需要帐户。将Base上的USDC发送到钱包地址，将tx哈希作为 payment_proof.

{
  "error": "Payment required",
  "x402": {
    "network": "base",
    "token": "USDC",
    "recipient": "0x9053FeDC90c1BCB4a8Cf708DdB426aB02430d6ad",
    "amount_usdc": 0.05
  }
}

______________________________________________________________________

用法（克劳德桌面/MCP客户端）

{
  "mcpServers": {
    "invoice-parser": {
      "url": "https://invoice-parser.plenitudo.ai/mcp?ref=readme"
    }
  }
}

______________________________________________________________________

建筑

server.py          — MCP server (6 tools: parse_invoice, parse_receipt,
                     extract_line_items, extract_totals, validate_invoice,
                     export_to_csv)
auth.py            — API key validation + usage tracking (SQLite)
x402.py            — x402 micropayment verification (USDC on Base)
config.py          — Environment variable loading
worker.js          — Cloudflare Worker (remote proxy for MCP traffic)
data/keys.db       — API key store (created at runtime)
data/usage.db      — Monthly usage counters (created at runtime)
logs/              — Structured log files
tests/             — Unit tests (mock Vision API, no real documents needed)

请求流

AI agent (Claude Desktop, Cursor, etc.)
    │
    │  MCP tool call (JSON-RPC over HTTP)
    ▼
Cloudflare Worker (worker.js)        ← optional remote proxy
    │
    │  Forwards to Railway deployment
    ▼
server.py (FastMCP, streamable HTTP)
    │
    ├── auth.py              validates API key / x402 proof
    ├── x402.py              verifies USDC transaction on Base
    └── Anthropic API        Claude Vision reads the document
            │
            ▼
        structured JSON → returned to agent

环境变量

______________________________________________________________________

部署（铁路）

1. 分叉此回购
2. 连接到铁路→ 新项目→ 从GitHub部署
3. 添加环境变量：

- ANTHROPIC_API_KEY -您的Anthropic API密钥 - INVOICEPARSER_DATA_DIR — /data - STRIPE_WEBHOOK_SECRET --来自Stripe仪表板 - STRIPE_PRO_PRICE_ID --来自Stripe仪表板

1. 在以下位置添加持久卷 /data
2. 部署

______________________________________________________________________

在本地运行（可选maxproxy路由）

如果您在与端口3456上的maxproxy实例相同的计算机上运行服务器，则可以通过它路由Claude Vision调用，而不是直接访问Anthropic API：

ANTHROPIC_API_KEY=maxproxy
ANTHROPIC_BASE_URL=http://localhost:3456

离开 ANTHROPIC_BASE_URL 在任何云/Railway部署中未设置（或为空）——这些环境无法访问本地代理。

______________________________________________________________________

贡献与安全

• 贡献.md --开发设置和PR指南
• 安全.md --负责任的披露政策
• 更改日志.md --版本历史记录

许可证

麻省理工学院 --版权所有©2025 Kenneth Nygren/Plenitudo AI