A document beside a laptop showing extracted fields, source references, checks, and review decisions

PDF and Word table converter

pdf-mcp

Turn document tables into Excel, CSV, or JSON.

Give it a PDF or Word file. It finds the tables, extracts the rows, and returns data your software or AI agent can use. Every value can point back to its source, and uncertain results are marked for review.

PDF or DOCX in Excel, CSV, or JSON out No OCR for scanned files

InputPDF and Word tables
OutputExcel, CSV, and JSON
Checksuncertain rows are flagged
Privacylocal mode uploads nothing

Simple browser app

Choose your computer. No Python or terminal.

Download the ZIP, unzip it, and double-click the app. It opens in your browser and processes documents only on your computer.

Community beta, currently unsigned. Your computer may ask you to confirm before opening it. See all v0.4.0 downloads.

A simple example

A PDF table goes in. Clean spreadsheet rows come out.

This sample invoice contains an item, quantity, and price table. pdf-mcp extracts all three rows and creates an Excel workbook plus CSV or JSON data. It also records where each value appeared in the document.

Input
1 PDF invoice with 3 line items
Output
The same 3 rows in Excel, CSV, or JSON
Result
Accepted, with no issues found
sample-invoice-output.xlsx Accepted
ItemQuantityUnit priceSource
Widget312.50Page 1, row 1
Gadget140.00Page 1, row 2
Bolt100.25Page 1, row 3
3rows extracted
0issues found
Acceptedready to use

Real output from the included synthetic invoice example.

Access

Use it locally, through an API, or with our help.

The downloadable browser app works without Python or an agent. Hosted API access and document setup services are request-only.

Simple browser app

For anyone converting a document

Available now

Download it, open it, and drop in a PDF or Word file. It runs locally, so nothing is uploaded.

Choose a download

Hosted API beta

Free during beta

Applications open

Send a redacted file to an API and receive the extracted data without running the tool yourself.

Applications are open. Invitations begin after endpoint deployment. Send no documents by email.

Apply for access

Document setup pilot

From EUR 19

Requests open

We test sample documents, configure the fields you need, and show what works and what needs review.

View pilot scopes

Hosted usage records exclude filenames, document contents, extracted text, and table cells. Read the privacy boundary.

What has been tested

Clean tables, uncertain rows, and wrong documents.

We created 18 fictional water-report documents to test all three outcomes. These tests show that the software behaves consistently on those examples. They are not customer evidence and do not predict accuracy on your documents.

See every test document and result
18
fictional test documents
12
used while building
6
kept as separate tests
3
possible outcomes

How reliable extraction is set up

Show it what to find, then test the answers.

  1. 01Name the fields

    Specify the columns you need, such as item, quantity, price, or sample ID.

  2. 02Extract the rows

    Read the table and keep a reference to the source of every value.

  3. 03Check the answers

    Compare the extracted rows with answers approved by a person.

  4. 04Handle uncertainty

    Use clean results and hold uncertain or unsupported documents for review.

Good fit

Best for document types you receive again and again.

  • Lab reports and certificates of analysis
  • Invoices, vendor statements, and operational reports
  • Tables that use the same columns across many files
  • Workflows where a person can check unusual results

Try it on your documents

Test one real document before committing.

Self-serve checkout remains disabled until business registration details are published.

Single-Document Feasibility Check

EUR 19 excluding VAT

1 document | 10 pages | 5 detected tables

  • Excel, CSV, and/or JSON output with source references
  • Notes explaining what was extracted and what may fail
  • One correction round within 7 days
  • Target delivery in 2 business days
Request feasibility check

Scope request by email. Do not attach the document.

If accepted documents cannot produce a usable versioned profile and evaluation baseline, choose a full refund or no-cost cancellation before manual rework.

Free local software, private customer data

Inspect the code. Keep your documents and test answers private.

The local pdf-mcp software is MIT licensed and sends no usage data. Settings made for a customer's document type and the correct answers used for testing stay private unless that customer chooses to share them. The hosted beta is a separate service.