AI Invoice Data Extraction: How an AI Agent Turns Financial Documents into ERP-Ready JSON via API and MCP

RUKN has launched the first agent in its new Agents for ERP lineup: Bill Agent, an AI Agent that reads a photographed, scanned, or PDF bill and returns clean, ERP-ready JSON — vendor, line items, taxes, and totals, with the math already checked. You can call it two ways: as a REST API from your own ERP, accounting system, or backend, or as an MCP server that gives AI assistants like Claude a tool to read bills directly inside a conversation or an automated workflow. It's live today at bill.rukn-software.com, currently offered as an early trial ahead of being built directly into RUKN ERP — and, the team hopes, into ERPNext and Odoo as well.

If you've ever handed an accountant a shoebox of receipts, or watched a finance team retype the same invoice fields for the third time this week, this is built for exactly that moment. Here's what it actually does, how the API and the MCP server work, and how to try it yourself.

Key takeaways

• Bill Agent turns a photo, scan, or PDF of a bill into structured JSON — vendor, customer, dates, line items, taxes and totals — in about 30 seconds for a one-page document.
• It checks its own math. Every bill gets an arithmetic validation pass and a confidence score per field, so you know exactly what to trust and what to glance at before it hits your books.
• You can plug it in two ways: a REST API for your own software, or an MCP server that lets AI agents such as Claude call it as a tool, no separate integration required.
• It reads Arabic, English, and Urdu, with number formats and tax rules tuned for Oman, Saudi Arabia, the UAE, and Pakistan — a gap most invoice OCR tools built for Western markets don't cover well.
• Pricing is transparent and small: roughly a cent per page, with a free daily allowance and hard spending caps, so testing it costs close to nothing.

Curious what an AI Agent does with your messiest invoice? Try Bill Agent free right now — no setup, just a photo.

Table of contents

1. What AI invoice data extraction actually means
2. Meet Bill Agent: RUKN's first step into agentic ERP
3. From photo to verified JSON: how the agent reads a bill
4. Built for Arabic, English and Urdu: solving Gulf invoice OCR properly
5. Two ways to plug it in: the REST API and the MCP server
6. Why an MCP server changes how you automate invoices
7. Trusting the numbers: how the agent checks its own math
8. What it costs to automate invoice data entry
9. Bill Agent in Oman, Saudi Arabia and the UAE
10. What's next: inside RUKN ERP, ERPNext, and beyond
11. How to try Bill Agent today
12. Frequently asked questions
13. Sources

What AI invoice data extraction actually means

AI invoice data extraction is the general name for software that reads a bill — a photo, a scan, or a PDF — and turns it into structured data your systems can actually use, instead of a flat image a human has to retype. It's a specific, more advanced case of a broader field called intelligent document processing, or Document AI: software that doesn't just recognize characters on a page (that's plain OCR) but understands what those characters mean — which number is the total, which date is due, which line is a discount.

The reason this matters for accounts payable automation is simple. Manual invoice entry is repetitive, error-prone, and expensive in staff time that could be spent on higher-value work. AI-powered accounting automation replaces that retyping with a read-once, use-everywhere flow: the invoice is photographed or forwarded once, and the data flows straight into your accounting system, your ERP, or your spreadsheet, already checked and structured.

Where Bill Agent pushes past standard invoice OCR is in what happens after the reading: it validates the arithmetic, flags anything uncertain for a human to check, and hands the result to you in two different ways depending on whether you're a developer or you're working through an AI assistant. That's what the rest of this guide walks through.

Meet Bill Agent: RUKN's first step into agentic ERP

RUKN builds ERP software for Gulf businesses, and Bill Agent is the company's first entry into what it's calling Agents for ERP — a lineup of focused AI agents meant to handle specific, well-defined jobs inside how a business runs, rather than one giant AI feature bolted onto everything at once. Bill Agent's job is reading bills: photographs, scans, and PDFs, in Arabic, English, or Urdu, from Oman, Saudi Arabia, the UAE, or Pakistan.

It's live today at its own address, bill.rukn-software.com, as a standalone product you can try immediately — no RUKN ERP account required. The team has described this as a trial launch: Bill Agent works fully today, with real API keys and real pricing, while RUKN gathers feedback ahead of building it directly into RUKN ERP itself, and potentially into ERPNext and Odoo afterward. If you want a closer look at how this fits into RUKN's wider Accounting & Finance capability, that's covered further down — but Bill Agent itself is worth understanding on its own first.

This is also, as far as we've found while researching this piece, one of the first invoice-reading agents built specifically for the Gulf that ships with a native MCP server on day one — most tools in this space, in the region and globally, only offer a REST API. More on why that distinction matters shortly.

From photo to verified JSON: how the agent reads a bill

The flow is intentionally simple. You send a bill — a phone photo, a scan, or a PDF, in JPEG, PNG, WebP, HEIC, or PDF format — and within about 30 seconds for a single page, you get back one JSON document with everything a bill normally contains: vendor and customer details, invoice and due dates, every line item, tax breakdowns, and totals.

A few details matter more than they might sound. First, every value is cleaned before it reaches you: amounts come back as actual numbers, dates in a consistent format, countries and currencies as standard codes — regardless of whether the original bill used Arabic numerals, a local date format, or a currency symbol. Second, a missing field comes back as genuinely empty rather than guessed; the agent doesn't invent a number that wasn't printed on the document. Third, longer documents — up to 30 pages in one request — are read in the background as a job you can poll or get a webhook for, so a long invoice doesn't hold your integration hostage waiting on a slow response.

There's also a specific answer for documents that aren't standard invoices. A handwritten shop ledger, common across small traders in the Gulf, comes back structured as a list of entries rather than forced into an invoice-shaped template it doesn't fit. A utility bill returns its own fields, like meter readings and account numbers, that a plain invoice extractor would have no place to put. This is document data extraction built to match the reality of Gulf paperwork, not just Western-style single-page invoices.

Built for Arabic, English and Urdu: solving Gulf invoice OCR properly

Arabic OCR is a genuinely hard problem, and it's one that a lot of invoice automation tools built primarily for Western markets simply don't solve well: right-to-left script, joined letters, mixed Arabic-English lines on the same invoice, and — specific to this region — Eastern Arabic numerals sitting next to Western digits on the same page, sometimes in the same line.

Bill Agent reads Arabic, English, and Urdu, including bilingual invoices where a single line mixes two of the three. That third language, Urdu, is worth pausing on: it's a deliberate, practical choice for a Gulf market where a meaningful share of small-business suppliers and tradespeople write receipts in Urdu, not just Arabic or English — a case most invoice OCR tools, Gulf-focused or global, don't handle at all.

The reading is also tuned to local numbers and tax rules rather than a single generic template: Oman, the UAE, Saudi Arabia, and Pakistan each have their own currency, decimal, and VAT conventions, and the agent applies the right one automatically or from a hint you provide. That's the difference between an OCR tool that transcribes characters and one that actually understands what it's reading.

Two ways to plug it in: the REST API and the MCP server

This is the part that separates Bill Agent from a simple "upload a PDF, download an Excel file" tool: the same underlying agent is exposed two different ways, for two different kinds of user.

The REST API — for your ERP, accounting system, or backend

If you're a developer connecting invoice reading to your own software — an ERP, an accounting platform, a mobile app, or an internal tool — the REST API is a straightforward POST request. Send the file, get back the JSON document in the same request for a single page, or queue it as a background job with a webhook for longer documents. Every call carries an API key, and the full schema is published openly, so you can generate types for your own codebase without needing a key just to see the shape of the data. This is the same integration pattern any invoice API or document processing API uses, whether you're a two-person startup or a large accounts payable team.

The MCP server — give Claude and other AI agents a tool to read bills

The second option is newer territory for most finance software, and it's the more interesting one: Bill Agent runs a Model Context Protocol server, meaning an AI assistant like Claude can call it directly as a tool, in the middle of a conversation or as part of an automated agent workflow, using the same API key and the same underlying reading engine as the REST API. Practically, that means someone can hand an assistant a bill and ask it to extract and file the details, and the assistant calls Bill Agent's MCP tools to actually do the reading — rather than trying to read the image itself, less reliably, with no arithmetic check and no structured schema to fall back on.

For a developer building on top of an AI agent framework, this is what "MCP ERP" or "AI Agent for ERP" concretely looks like in practice: not a chatbot bolted onto a dashboard, but a real tool an agent can call, with the same guarantees — schema, confidence scores, arithmetic validation — as calling the API directly.

Why an MCP server changes how you automate invoices

If you're not deep in AI tooling, here's the short version of why this matters. Model Context Protocol, or MCP, is an open standard that lets AI assistants connect to external tools and data sources in a consistent way — instead of every company building a one-off, custom integration for every AI product a business might use, an MCP server exposes a tool once, and any MCP-compatible assistant can call it.

For invoice automation specifically, this means Bill Agent doesn't just serve one product's users. Any team already working inside an AI-agent-based workflow — using Claude Code, or another MCP-compatible agent to help run parts of their business — can add bill reading as a capability in minutes, the same way they'd add any other tool, rather than building a custom OCR integration from scratch. That's the practical meaning behind "Agentic ERP" and "Agentic AI ERP": ERP-adjacent capabilities exposed as tools an agent can pick up and use, not a separate application someone has to log into.

This is also, genuinely, still rare. Most of the invoice-reading tools built for this region — and most of the well-known global players — offer a REST API and stop there. An MCP server on day one, alongside the API, puts Bill Agent ahead of that curve rather than catching up to it later.

Trusting the numbers: how the agent checks its own math

The single biggest risk in any automated invoice reading is quiet, confident wrongness — a number that looks right, gets filed, and turns out not to match what was actually on the page. Bill Agent is built around not letting that happen silently.

Every bill goes through an arithmetic validation pass: line items are checked against the subtotal, tax is recalculated against the stated rate, and the final total is checked against what was printed. If something doesn't add up, the response says exactly what and by how much — not just a vague "please review" flag. Separately, every individual field carries its own confidence score, and anything read with lower confidence than your threshold is listed explicitly, so a reviewer's attention goes to the two unclear fields on a bill, not the whole document.

This is the difference between "AI read this invoice" and "AI read this invoice, and here's precisely what it's sure of, what it's not, and what didn't add up" — the second is the one worth automating around, because it tells you when to trust it and when to look closer, instead of asking you to trust it blindly every time.

What it costs to automate invoice data entry

Invoice OCR pricing across the market varies enormously, and it's worth knowing the range before judging any single number. Established platforms in this space commonly charge anywhere from roughly thirty cents a page to several dollars per document once volume and support are factored in, and some enterprise-focused platforms start in the thousands of dollars a month before you've processed a single invoice.

Bill Agent's published rate sits at roughly one cent per page, with a genuinely free tier for anything that isn't actually a document (a blank or irrelevant photo costs nothing), and hard daily and monthly spending caps on new accounts so testing it can't run away on you. That combination — a per-page cost measured in cents rather than dollars, with no subscription required to start — is a meaningfully different position from most of the invoice automation tools competing for the same job, and it's a large part of why trying it costs close to nothing.

Bill Agent in Oman, Saudi Arabia and the UAE

Invoice automation and AI document processing are genuinely active categories across the Gulf right now, largely driven by the wave of e-invoicing mandates moving through the region — Saudi Arabia's ZATCA requirements, the UAE's e-invoicing rollout, and Oman's own Fawtara system, which we've covered in detail in our Fawtara e-invoicing guide and our guide to ZATCA-compliant software in Saudi Arabia. A fair number of AI-powered invoice tools already market themselves around "ZATCA AI" specifically, reading Saudi tax invoices and their embedded QR codes.

Bill Agent's positioning is a step earlier and a step broader in that picture: rather than only validating an already-compliant e-invoice, it reads the messy paperwork that arrives before any of that — a handwritten receipt, a supplier's paper invoice, a shop ledger — and turns it into the same clean, structured data an e-invoicing or accounting system needs downstream. For a business already thinking about e-invoicing readiness, feeding a system with cleanly extracted, verified data from day one is a smaller leap than fixing years of inconsistent manual entry retroactively.

Oman in particular is worth calling out on its own: most of the AI invoice processing and document AI activity we found while researching this piece is concentrated in Saudi Arabia and the UAE, with comparatively little built specifically for Oman. A Muscat-headquartered ERP company shipping an Arabic-first AI agent, with Oman named explicitly among its supported countries, fills a gap that's mostly empty right now.

What's next: inside RUKN ERP, ERPNext, and beyond

Bill Agent today stands alone, reachable through its own web app, its REST API, and its MCP server, independent of whether you use RUKN ERP at all. The team's stated plan is to bring it directly inside RUKN ERP once the trial period wraps up — meaning a bill photographed today could, in a future version, flow straight into your accounting entries without ever touching a separate app or API call — and, the team hopes, to make it available within ERPNext and possibly Odoo as well, extending the same reading engine to businesses running those platforms.

None of that is live yet, and no firm date has been set — this is an early trial specifically meant to shape that next step, not a finished roadmap. If native integration into RUKN's Accounting & Finance and Compliance & Controls capabilities is what you're actually waiting for, the most useful thing you can do today is try Bill Agent as it stands and tell the team what worked and what didn't — that feedback is explicitly what this trial period is for.

How to try Bill Agent today

There are two ways in, depending on what you want to test.

If you just want to see it work, open Bill Agent and try your first bill — no setup, no account required to start, just a photo of any bill, receipt, or invoice you have on hand. Give it your worst one: handwritten, faded, in two languages at once. That's exactly the case it's built for.

If you're a developer and want to integrate it — into an ERP, an accounting tool, an internal system, or an AI agent workflow — the developer documentation covers the REST API and the MCP server in full, including the published JSON schema, so you can see exactly what comes back before writing a line of code.

And if what you're really interested in is where this is heading — Bill Agent built directly into a full ERP system alongside accounting, sales, purchasing, and inventory — get in touch with the RUKN team and mention Bill Agent specifically. Early interest is exactly what shapes which platform gets it next.

Frequently asked questions

What is Bill Agent? Bill Agent is an AI Agent built by RUKN that reads photographed, scanned, or PDF bills and returns clean, structured JSON — vendor, line items, taxes, and totals — with the arithmetic already checked. It's the first agent in RUKN's Agents for ERP lineup, live today at bill.rukn-software.com.

How is this different from regular invoice OCR? Standard OCR reads characters on a page. Bill Agent goes further: it understands which number is the total, which date is due, and which line is a discount, then validates the math and flags anything uncertain for review — rather than just handing back raw, unstructured text.

What languages and countries does it support? Arabic, English, and Urdu, including mixed-language invoices, with number formats and tax rules tuned for Oman, Saudi Arabia, the UAE, and Pakistan.

What is an MCP server, and why does it matter for invoices? MCP, or Model Context Protocol, is an open standard that lets AI assistants like Claude call external tools consistently. Because Bill Agent runs an MCP server, an AI assistant can read a bill directly as a tool inside a conversation or automated workflow, using the same engine and guarantees as the REST API — without a custom, one-off integration.

Can I use this without any technical setup? Yes. The web app at bill.rukn-software.com works by simply uploading a photo, scan, or PDF — no API key or integration required to try it. The API and MCP server are there when you're ready to connect it to your own systems.

How much does it cost? Roughly a cent per page, with a free allowance for non-document uploads and hard daily and monthly spending caps on new accounts, so testing costs close to nothing.

Will this be part of RUKN's full ERP system? That's the plan. RUKN intends to build Bill Agent directly into RUKN ERP once this trial period ends, and hopes to extend it to ERPNext and potentially Odoo as well. No firm date has been announced yet.

Is my data safe when I upload a bill? Bill history is private to your own account, can be reviewed and corrected later, and can be deleted at any time. Review the product's own documentation for full details before sending sensitive documents.

Try it yourself

The fastest way to understand what an AI Agent can actually do with your invoices isn't reading about it — it's handing it one.

Try Bill Agent free · Read the developer docs (API & MCP) · Ask RUKN about Bill Agent inside your ERP

Sources and how we checked

Product details in this guide were checked directly against Bill Agent's own site and developer documentation on 26 September 2026: the Bill Agent homepage and its developer documentation, covering the REST API, the MCP server, pricing, and supported languages and countries.

Market and competitive context: Docsumo's invoice OCR software comparison; Extend's Nanonets pricing review; invoiceocr.ai's 9-tool invoice OCR comparison; and general observation of Gulf-market AI invoice tools (Raqeem AI, WinTrack, TronStride/Aigentrix, Skyline, Sanad.ai) via their own public sites, checked the same day, referenced here only to describe the competitive landscape and not to make claims about those products' accuracy or compliance status.

Bill Agent is a trial-stage product; features, pricing, and the API and MCP interfaces described here may change as RUKN develops it further. Check the product's own documentation for the current state before building anything on top of it.