PDF, DOCX & XLSX to Text

What does PDF, DOCX & XLSX to Text do?

PDF, DOCX & XLSX to Text converts PDF, Word and Excel documents to text. Give it document links, including scanned PDFs and images (read with OCR); you get the plain text, the same content as Markdown, tables as rows, and the document's metadata.

PDF text is read with a fast engine by default. Turn on Extract PDF tables when you need a PDF's tables as rows and as Markdown tables; Word and Excel tables always come back.

Markdown keeps Word headings and lists, turns tables into Markdown tables and marks each PDF page, ready to chunk for RAG or paste into an LLM prompt. The file type is read from the file itself, so links like .../download?id=123 work as long as they lead straight to the file.

Try it in one click: the input comes pre-filled with three small US government documents: IRS Form W-9 (PDF, 6 pages), a USPTO patent application template (DOCX) and a BLS survey template (XLSX, 3 sheets). That's 3 documents, about $0.015 (3 × $0.005, plus $0.00005 for the run start). Then replace them with your own document links.

What data does PDF, DOCX & XLSX to Text return?

Field	Example	Notes
text	Form W-9\n(Rev. March 2024)\n...	The plain text, in reading order.
markdown	<!-- page 1 -->\n\nForm W-9\n...	Headings, lists and tables as Markdown; PDF pages marked.
tables	[{"page": 1, "sheet": null, "index": 1, "rows": [[...]]}]	Tables as rows: Word and Excel always, PDF with Extract PDF tables on; Excel cells keep numbers and true/false.
format	pdf	pdf, docx, xlsx, or the image format.
title, author	Form W-9 (Rev. March 2024), SE:W:CAR:MP	From the document's metadata; null if none.
createdAt, modifiedAt	2024-03-06T13:18:13Z
pageCount, pagesExtracted	6, 6	PDF pages (DOCX: what Word saved; null for Excel).
sheetNames	null	Excel only.
wordCount, tableCount	6272, 0
ocrPages, ocrConfidence	[], null	Pages read with OCR (the ones you pay for) and how sure OCR was.
fileName, fileSizeBytes	fw9.pdf, 140815
truncated, truncatedReasons	false, []	Whether part of the document was left out, and why.

One result per document. The full list is under Output.

How much does it cost to convert PDF, Word and Excel to text?

You pay per document converted: $5.00 per 1,000 documents (up to 100 pages each), plus $0.30 per 1,000 table pages (PDF pages read with Extract PDF tables on), plus $10.00 per 1,000 OCR pages (each scanned PDF page or image read with OCR that had text), plus $0.00005 each time a run starts.

PDF tables are opt-in, and charged per page, because they are what costs to run. Finding tables means reading every character and line on the page with a layout-aware parser: on a 100-page PDF that took 15-20 times the compute of reading its text. Plain text stays cheap for everyone who doesn't need rows; Word and Excel tables are always included in the document price.

• The example below: 3 documents × $0.005 = $0.015, plus the start fee. With Extract PDF tables on, the W-9's 6 pages add 6 × $0.0003 = $0.0018.
• A 20-page PDF with tables: $0.005 + 20 × $0.0003 = $0.011. A 10-page scanned PDF: one document and 10 OCR pages, $0.005 + $0.10 = $0.105.
• A month, for example: 200 documents: 200 × $0.005 = $1.00; 50 of them are 10-page PDFs read with tables, 500 table pages × $0.0003 = $0.15; 20 are 3-page scans, 60 OCR pages × $0.01 = $0.60: $1.75 plus the run starts.
• Caps: Max documents per run in the input, and Maximum cost per run in the run options. The run stops cleanly at whichever comes first. Table and OCR pages count against the maximum cost too: before a PDF is downloaded with tables on, the table pages it can need (up to Max pages per PDF) are reserved against what's left, and before a document's scanned pages are read, they are too (after the documents already in progress). If what's left doesn't cover them all, only the first pages are read and truncatedReasons says so. Each document is counted against the limit before it's downloaded (and the count given back if it fails), so a capped run never downloads documents or reads pages it can't return.

Never charged: pages with a text layer as OCR pages (they're never OCR'd), blank scanned pages, table pages when Extract PDF tables is off (or when a PDF's structure needs the text-only fallback reader), and documents that fail (password-protected without the right password, unsupported, too large, missing, no text even after OCR, none of the pages in Pages to read, or disallowed by the site's robots.txt), nor their pages. If you don't want OCR, turn off Read scanned pages with OCR: scanned PDFs and images are then reported, not charged.

The OCR page price is set from what OCR measurably costs to run: a scanned page is read at 300 DPI, and a dense page of text costs several times more compute than a typical photo. For single images, our Image to Text OCR actor is cheaper.

How to convert a PDF, Word or Excel file to text

1. Open PDF, DOCX & XLSX to Text and click Try for free (or Start if you're signed in).
2. Put direct links to your files in Document URLs, one per line.
3. Optional: set Max pages per PDF or Pages to read (e.g. 1-10), give a PDF password, turn Extract PDF tables or Read scanned pages with OCR on or off, and pick the Languages of scanned text.
4. Click Start, then open the Output tab and export as JSON, CSV or Excel.

Example: a PDF, a Word file and an Excel workbook

The pre-filled input:

{"urls": ["https://www.irs.gov/pub/irs-pdf/fw9.pdf",
          "https://www.uspto.gov/sites/default/files/documents/Initial-Filing-Template-August-2025.docx",
          "https://www.bls.gov/respondents/oes/template.xlsx"],
 "maxPagesPerDocument": 50, "extractTables": false, "ocr": true, "languages": ["eng"]}

The PDF's result (real output from a local run on 2026-09-25; text and markdown shortened, tables left out):

{
  "id": "60084aa5127e297e8b972cbe",
  "url": "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
  "fileName": "fw9.pdf",
  "format": "pdf",
  "fileSizeBytes": 140815,
  "title": "Form W-9 (Rev. March 2024)",
  "createdAt": "2024-03-06T13:18:13Z",
  "pageCount": 6,
  "pagesExtracted": 6,
  "wordCount": 6272,
  "tableCount": 0,
  "imageOnlyPages": [],
  "ocrPages": [],
  "text": "Form W-9\n(Rev. March 2024)\nRequest for Taxpayer\nIdentification Number and Certification\nDepartment...",
  "markdown": "<!-- page 1 -->\n\nForm W-9\n(Rev. March 2024)\nRequest for Taxpayer\n...",
  "truncated": false
}

The Excel workbook came back with its 3 sheet names and 3 tables (one per sheet, as rows); the Word file with its text (258 words). None of the three needed OCR. With Extract PDF tables on, the W-9 also comes back with its 4 tables (tableCount: 4), as rows and as Markdown tables in place.

Input

Field	What it does
Document URLs	Direct links to the files, one per line (PDF, DOCX or XLSX, or an image: PNG, JPEG, WebP, TIFF, GIF, BMP).
Max pages per PDF	Read at most this many pages of each PDF, from the first (default 50, at most 100). pageCount still gives the whole document's length.
Pages to read (PDF)	Only these pages of each PDF, e.g. 1-10 or 1-3, 7, 20- (20- = page 20 to the end). Max pages per PDF still applies.
PDF password	The password that opens your password-protected PDFs (stored encrypted; the same one is tried on each PDF that needs one).
Extract PDF tables	Find PDF tables and return them as rows and as Markdown tables (default off; slower, and each page read this way is a table page). Word and Excel tables always come back.
Read scanned pages with OCR	Read scanned PDF pages and images with OCR (default on). Each page read is an OCR page.
Languages	The language(s) of the scanned text, up to 4 (default English).
Max documents per run	Cap the number of documents converted in one run.

{
  "urls": [
    "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
    "https://example.com/files/price-list.xlsx"
  ],
  "maxPagesPerDocument": 50,
  "pages": "1-20",
  "extractTables": true,
  "ocr": true,
  "languages": ["eng"]
}

Output

One result per document. Fields a format doesn't have are null.

{
  "id": "60084aa5127e297e8b972cbe",
  "url": "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
  "finalUrl": "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
  "fileName": "fw9.pdf",
  "format": "pdf",
  "fileSizeBytes": 140815,
  "title": "Form W-9 (Rev. March 2024)",
  "author": "SE:W:CAR:MP",
  "createdAt": "2024-03-06T13:18:13Z",
  "modifiedAt": "2024-03-06T13:18:13Z",
  "pageCount": 6,
  "pagesExtracted": 6,
  "sheetNames": null,
  "wordCount": 6279,
  "tableCount": 4,
  "imageOnlyPages": [],
  "ocrPages": [],
  "ocrConfidence": null,
  "text": "W-9 Request for Taxpayer ...",
  "markdown": "<!-- page 1 -->\n\nW-9 Request for Taxpayer ...",
  "tables": [
    {"page": 1, "sheet": null, "index": 1, "rows": [["...", "..."]]}
  ],
  "truncated": false,
  "truncatedReasons": [],
  "scrapedAt": "2026-09-24T20:32:14Z"
}

• id is stable across runs (it comes from the URL), so you can use it to deduplicate or as a vector-store key.
• tables: page is set for PDFs, sheet for Excel; index counts tables within the document. PDF and Word cells are text; Excel cells keep numbers and true/false, and dates are ISO 8601 text. PDF tables are only looked for with Extract PDF tables on (the example above had it on).
• pageCount: PDF pages in the whole document; pagesExtracted: how many of them were read. Word documents have no fixed pages (they depend on fonts and paper), so for DOCX this is the count Word saved in the file, when it did. null for Excel.
• imageOnlyPages: PDF pages that are only pictures (scanned). With OCR on, their text is read by OCR.
• ocrPages: the pages whose text came from OCR (these are the OCR pages you pay for; an image is page 1), and ocrConfidence Tesseract's mean word confidence over them, 0-100 (null when no page needed OCR). A low confidence usually means a blurry or low-resolution scan, or the wrong language.
• format: pdf, docx, xlsx, or for images png, jpeg, webp, tiff, gif or bmp.
• truncated / truncatedReasons: whether part of the document was left out, and why (see the limits below).

What it reads, and what it doesn't

• PDF: text and metadata. By default the text is read with PDFium (the engine inside Chrome's PDF viewer): fast, column by column on multi-column pages, with words hyphenated at line ends joined again. A table's cells come back as text in the order the PDF draws them, which for dense tables can be cell by cell rather than row by row. With Extract PDF tables on, the PDF is read with pdfplumber instead: tables drawn with ruling lines come back as rows and as Markdown tables where they sit, and the text is read line by line across the page (on two-column pages, that interleaves the columns). Tables drawn without lines, and tables on scanned pages, are returned as text. A PDF that only restricts printing or copying opens normally, as in any PDF viewer; one that needs a password to open is read with the PDF password you give, and otherwise reported as password-protected. Scanned pages are read with OCR at 300 DPI (Tesseract 5, LSTM engine), in the languages you pick.
• Images (PNG, JPEG, WebP, TIFF, GIF, BMP): read with OCR, the same way as our Image to Text OCR actor: one page per file (a multi-page TIFF is refused), the first frame of an animation, phone photos turned upright, up to 50 megapixels.
• Word (.docx): paragraphs, headings, lists, tables and content controls, in document order. Headers, footers, comments and footnotes are left out. Old .doc files aren't supported.
• Excel (.xlsx): every worksheet's values. A formula gives the result Excel last saved; a workbook written by a program that never calculated its formulas has no saved result, so those cells are empty. Old .xls files aren't supported.

Run it on a schedule, or from your own code

1. Save your input as a task (Save as a new task, top right of the actor page) and add it to a schedule (Console → Schedules → Create new): for example daily, if the same links are republished with new content (a price list, a monthly report). There's no "only changed" mode: every run converts, and charges for, every document you list. Each result's id stays the same from run to run for the same URL, so you can replace the old version.
2. Collect results: download the dataset as JSON, CSV or Excel; fetch the latest run's results from the API (GET https://api.apify.com/v2/actor-tasks/<task id>/runs/last/dataset/items?status=SUCCEEDED&format=csv, with your API token); let a webhook tell your system when a run succeeds; or connect it to Make, Zapier or n8n through Apify's integrations.

Can I use PDF, DOCX & XLSX to Text from an AI agent (MCP)?

Yes, through Apify's MCP server: add https://mcp.apify.com?tools=humble-echidna/document-to-text to your MCP client (or let the agent find it with the server's actor search). The agent passes document links, e.g. {"urls": ["https://example.com/report.pdf"], "maxPagesPerDocument": 10}, and reads markdown back.

Who it's for

Teams that receive documents as links (supplier price lists, filings, reports, forms, contracts published online) and need their text and tables in a database, a spreadsheet or an LLM pipeline, without opening each file.

Why this one?

• Three formats, one output shape. PDF, Word and Excel come back with the same fields, so one pipeline handles all of them.
• Fast text, tables when you want them. Plain PDF text is read with a fast engine; turn on Extract PDF tables and a PDF table sits where it is on the page, as a Markdown table, instead of being flattened into lines of text, and comes back as rows too. Word and Excel tables always do (Excel numbers and true/false values keep their type).
• Page ranges and passwords. Read only the pages you need (1-10, 20-), and open password-protected PDFs with your password.
• Scanned pages get text too. A PDF page that is only a picture is read with OCR, and its text goes where the page is, between the pages that already had text. Only those pages go through OCR, and only pages where OCR finds text are charged as OCR pages; a blank scanned page costs nothing.
• Honest about what it can't read. Password-protected PDFs are reported as such and not charged, instead of coming back as an empty result. The same goes for old .doc and .xls files, PowerPoint, links that lead to a web page instead of a file, and missing files.
• No minimum. The start fee is $0.00005, so converting a single document costs about half a cent.
• Reliable. Each document is parsed in its own process with a time limit and a memory limit, and each OCR page is read in its own process with its own limits, so one broken or huge file (or page) fails alone and never stops the rest of your run. The run log and the RUN_STATS record say exactly which document had a problem and why.

Limits

• 50 MB per file, and about 1 GB of downloads per run (checked as each download starts). Each file gets 2 minutes to download, retries included; a server slower than that (about 420 KB/s for a 50 MB file) is reported, not charged.
• PDF: up to 100 pages per document (default 50), from the first page or from Pages to read.
• Up to 800,000 characters of text (and of Markdown) per document, and 800,000 characters of table cells.
• Excel: up to 20,000 rows per sheet, 500 columns and 500,000 cells per workbook.
• Word and Excel files that unpack to more than 300 MB are not opened, and neither is anything shaped like a ZIP bomb.
• Each document is parsed in its own process with a memory cap and a time limit: 1 minute for a PDF read without tables (it takes seconds even for 100 pages), 3 minutes with Extract PDF tables on and for Word, Excel and images, and 3 minutes per OCR page. With tables on, a dense 100-page PDF can take about 2-3 minutes at the default 2 GB; for long, table-heavy PDFs run with 4 GB.
• OCR is the slow part, and it's limited by CPU, which Apify gives in proportion to memory (one core per 4 GB). A dense scanned page (a full tax form, about 1,000 words) took about 8 seconds on one core in our tests: about 16 seconds per page with the default 2 GB, and a 20-page scan in about 5.5 minutes. OCR used less than 200 MB of memory, so more memory only makes it faster, up to the 4 GB maximum (one full core).

robots.txt and politeness

• Before downloading a file, including every redirect hop, it checks that site's robots.txt and skips files it disallows. robots.txt is read once per site per run. Skipped files are listed in the log and in RUN_STATS (blockedByRobots), and are never charged.
• It identifies itself honestly, with the user agent HumbleEchidnaApify/1.0 (+https://apify.com/humble-echidna). Site owners can allow or block it by name: robots.txt rules written for User-agent: HumbleEchidnaApify apply to it (otherwise the * rules do).
• It also honours opt-outs from AI use. Its output is meant for LLM and RAG pipelines, so when a site's robots.txt disallows a file for any of the well-known AI crawlers (GPTBot, CCBot, ClaudeBot, Google-Extended, PerplexityBot and the like), the file isn't downloaded. It's reported as opted out of AI use (optedOutOfAI) and not charged.
• It honours Crawl-delay and Retry-After, and never downloads more than 2 files at a time from one site.
• It only fetches public web addresses on the standard ports (80 and 443). It doesn't log in, and doesn't try to get around paywalls, logins or bot protection.

FAQ

Can it read scanned PDFs or images?

Yes, with OCR (Tesseract), in 32 languages: pick the ones in your documents under Languages. Each page read by OCR is charged as an OCR page. Handwriting, very small print and low-resolution scans read poorly; ocrConfidence shows how sure OCR was.

Languages: English, German, French, Spanish, Italian, Portuguese, Dutch, Polish, Czech, Slovak, Swedish, Danish, Norwegian, Finnish, Hungarian, Romanian, Turkish, Greek, Russian, Ukrainian, Bulgarian, Arabic, Hebrew, Hindi, Bengali, Thai, Vietnamese, Indonesian, Japanese, Chinese (Simplified), Chinese (Traditional) and Korean.

Can it open password-protected PDFs?

Yes, if you give the password in PDF password (it's stored encrypted and only used to open the files inside the run). Without it, or with a wrong one, they're reported as password-protected and not charged.

Should I turn on Extract PDF tables?

Only if you need a PDF's tables as rows or as Markdown tables. Without it, the table's text is still there, and the rest of the PDF is read faster and, on two-column pages, in a better reading order. Word and Excel tables come back either way.

Why does my link return "a web page, not a document"?

Some sites put a download page or a viewer in front of the file. Open the link in a browser, copy the address of the file itself, and use that.

Something that used to work now fails. Why?

The run log names the document and what went wrong, and every other document in the run is unaffected. Please open an issue with the input you used.

Is it legal to convert documents to text?

It only downloads the files you give it, from public web addresses, without logging in, and it follows each site's robots.txt. Make sure you have the right to process the documents you send it.

Related actors

Actor	Use it when
Image to Text OCR	You have single images (photos, screenshots, one-page scans): $4.00 per 1,000 images, with line and word positions.
Website & Page to Markdown for RAG and LLMs	You have web pages (HTML) rather than files.
Academic Papers Search: OpenAlex + Crossref	You want papers: it finds them with their open-access pdfUrl, which you can pass here to get the text, where the host's robots.txt allows it.

Feedback and support

Found a bug, or a document that converts badly? Open an issue on the Issues tab with the link you used.

Versions

Current version: 1.3. See the Changelog tab for what changed in each version.