PDF & Image to Text OCR for AI Agents - No Login

Turn public PDF and image URLs into clean text or Markdown. The Actor reads the PDF text layer when a page has one, and runs Tesseract OCR on scanned pages and images, page by page. No login, no API key. A run with one text PDF costs about $0.0075; each OCR page adds $0.012.

Add one URL and click Start: a 6-page text PDF plus one image takes about 4 seconds and costs about $0.02. You get one row per document (or one row per page for RAG), with the reading method and OCR confidence for every page.

Table of contents

• What it does
• What data you get
• Quick start
• Input
• Output
• Pricing
• Use cases
• Scheduling and monitoring
• API, export and integrations
• Limitations and partial results
• Troubleshooting
• FAQ
• Related Actors
• Support

What it does

You give public http(s) URLs of PDF files or images. For each URL the Actor:

1. Downloads the file and checks its real type from the file content (not from the URL).
2. Reads every selected page of a PDF:
   • Text layer: if the page has real text, it is read directly. This is fast and exact.
   • OCR: if the page is a scan (no usable text), it is rendered at 300 DPI and read with Tesseract OCR.
3. Reads images (PNG, JPG, TIFF, GIF, BMP, WEBP) with OCR.
4. Saves the text as plain text or Markdown to the Dataset, with page numbers, method (text or ocr) and OCR confidence.
5. Deletes the downloaded file. Documents and page images are never stored.

Good fit

• AI agents and RAG pipelines: get clean text from a PDF link, one row per page, ready for chunking.
• Scanned archives and forms: read old scans that have no text layer.
• Mixed batches: text PDFs, scans and photos of documents in one run, with clear status per file.

What data you get

For each URL you get one row with the text, the page count, how many pages were read from the text layer and how many with OCR, the PDF title and metadata, and a clear status with an error code when something went wrong. In page mode you get one row per page instead. See Output for all fields.

Quick start

1. Open the Input tab.
2. Keep the two prefilled urls (an IRS W-9 form and a small test image), or add your own PDF or image URL.
3. Keep OCR mode auto and OCR languages English.
4. Click Start.
5. Open the Documents view in the Output tab and check status, textPages, ocrPages and text.

Expected result: two rows with status: complete. The W-9 has 6 text-layer pages; the image has 1 OCR page.

Input

Input fields

Field	Plain name	Type	Required	Default / prefill	Allowed values	Effect	Example	Recommendation
urls	PDF or image URLs	array of strings	yes	prefill: W-9 PDF + eurotext.png	1–100 public http(s) URLs	Files to read. Duplicates are read once. Objects like {"url": "..."} are also accepted.	["https://www.irs.gov/pub/irs-pdf/fw9.pdf"]	Use direct file links, not viewer pages.
ocr	OCR mode	string	no	auto	auto, always, never	auto: OCR only pages without usable text. always: OCR every page. never: text layer only; images fail with OCR_DISABLED.	"auto"	Keep auto. Use always only when the text layer is broken.
languages	OCR languages	array of strings	no	["eng"]	up to 3 of: eng, deu, fra, spa, ita, por, nld, pol, ces, swe, dan, nor, fin, tur, rus, ukr, ara, hin, vie, ind	Languages for OCR. Each OCR page is charged once per language.	["eng", "deu"]	Pick only the languages the document uses.
pages	Pages	string	no	empty (all pages)	ranges like 1-3, 5, 10-, up to 50 ranges	Same page selection for every document. 10- means page 10 to the end.	"1-3"	Use it to skip covers or appendices.
maxPages	Max pages per document	integer	no	50	1–300	Upper limit after the page selection. Extra pages are skipped, not charged, and listed in warnings.	100	Raise it for long text PDFs.
outputFormat	Text format	string	no	text	text, markdown	markdown re-flows paragraphs, keeps bullet and numbered lists, makes short ALL-CAPS lines headings and adds page markers.	"markdown"	Use markdown for LLM input.
outputMode	Rows	string	no	document	document, page	document: one row per document with a pages list. page: one row per page.	"page"	Use page for RAG chunking.
preserveLayout	Keep column layout	boolean	no	false	true, false	Keeps the physical layout of text-layer pages (columns, spacing). No effect on OCR pages.	true	Use for forms and invoices with aligned columns.
maxFileSizeMb	Max file size (MB)	integer	no	50	1–100	Larger files are not downloaded and get FILE_TOO_LARGE.	100	Raise only for big scanned books.

Empty urls stops the run with INVALID_INPUT. An API call with {} does not use the prefill.

Interacting settings

When	Input A	Input B	Effective behavior	Example	Cost/result effect
Page selection is long	pages	maxPages	pages is applied first, then maxPages cuts the rest	pages: "1-", maxPages: 10 → pages 1–10	Skipped pages are not charged
Several languages	languages	ocr	Languages are used only on OCR pages	["eng","fra"] on a text PDF	No OCR, no language charge
OCR is off	ocr: never	outputMode	A scan gives one row with no text: text: "" and pages with method: none in document mode, text: null and method: none in page mode	Scanned PDF	One document event: the PDF was opened and checked
Page rows	outputMode: page	text-layer pages	One document event per started 50 text-layer page rows	300 text pages → 6 events	See Pricing

Input examples

Read a text PDF and an image

{
  "urls": [
    "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
    "https://raw.githubusercontent.com/tesseract-ocr/tessdoc/main/images/eurotext.png"
  ]
}

Expected behavior: the W-9 is read from its text layer (6 pages); the image is read with English OCR. Expected output: 2 document rows with status, textPages, ocrPages, text. Cost: 1 document + 1 ocr-page + start ≈ $0.0195 on the Free plan.

Page rows in Markdown for RAG

{
  "urls": ["https://www.irs.gov/pub/irs-pdf/fw9.pdf"],
  "outputMode": "page",
  "outputFormat": "markdown",
  "pages": "1-3"
}

Expected behavior: pages 1–3 are read and saved as 3 rows. Expected output: pageNumber, method, charCount, text per row.

Output

Results go to the default Dataset (view Documents). The run summary goes to the key-value store record RUN_SUMMARY. You can export the Dataset as JSON, CSV, Excel or HTML.

Output example

Input: the first example above. Two rows; text and pages are trimmed.

[
  {
    "url": "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
    "finalUrl": "https://www.irs.gov/pub/irs-pdf/fw9.pdf",
    "status": "complete",
    "fileType": "pdf",
    "contentType": "application/pdf",
    "fileSizeBytes": 140815,
    "pageCount": 6,
    "pagesProcessed": 6,
    "pagesTruncated": false,
    "textPages": 6,
    "ocrPages": 0,
    "format": "text",
    "ocrLanguages": null,
    "charCount": 37820,
    "text": "W-9\n\nForm\n(Rev. March 2024)\nDepartment of the Treasury\nInternal Revenue Service\n\nRequest for Taxpayer\nIdentification Number and Certification…",
    "pages": [
      { "pageNumber": 1, "method": "text", "charCount": 5703, "ocrConfidence": null, "text": "W-9\n\nForm\n(Rev. March 2024)…", "error": null }
    ],
    "title": "Form W-9 (Rev. March 2024)",
    "metadata": { "subject": "Request for Taxpayer Identification Number and Certification", "creator": "Designer 6.5", "pdfVersion": "1.7" },
    "encrypted": false,
    "error": null,
    "warnings": [],
    "fetchedAt": "2026-10-13T05:23:31.200Z",
    "processedAt": "2026-10-13T05:23:31.254Z"
  },
  {
    "url": "https://raw.githubusercontent.com/tesseract-ocr/tessdoc/main/images/eurotext.png",
    "status": "complete",
    "fileType": "png",
    "pageCount": 1,
    "textPages": 0,
    "ocrPages": 1,
    "ocrLanguages": ["eng"],
    "charCount": 414,
    "text": "The (quick) [brown] {fox} jumps!\nOver the $43,456.78 <lazy> #90 dog\n& duck/goose, as 12.5% of E-mail\nfrom aspammer@website.com is spam.…",
    "pages": [
      { "pageNumber": 1, "method": "ocr", "charCount": 414, "ocrConfidence": 90.2, "text": "The (quick) [brown] {fox} jumps!…", "error": null }
    ],
    "title": null,
    "metadata": null,
    "error": null,
    "warnings": []
  }
]

OCR is not perfect: check ocrConfidence. Low values (below about 60) usually mean a blurry scan, the wrong language or handwriting.

Output fields

Field	Meaning	Type	Can be empty?	Example / enum	Notes
url	URL from your input	string	no	https://www.irs.gov/pub/irs-pdf/fw9.pdf	Kept exactly as given
finalUrl	URL after redirects	string	yes, when not downloaded	same as url
status	Result of this row	string	no	complete, partial, not_found, error	See the table below
fileType	Type found in the file content	string	yes	pdf, png, jpg, tiff, gif, bmp, webp
contentType	Content-Type sent by the server	string	yes	application/pdf	Document rows
fileSizeBytes	Downloaded size	integer	yes	140815	Document rows
pageCount	All pages in the file	integer	yes	6	1 for single images
pagesProcessed	Pages read with success	integer	no	6	Document rows
pagesTruncated	maxPages cut the selection	boolean	no	false	Document rows
textPages	Pages read from the text layer	integer	no	6	Document rows
ocrPages	Pages read with OCR	integer	no	0	Document rows
pageNumber	Page number, from 1	integer	page rows only	1	Page rows
method	How the page was read	string	page rows only	text, ocr, none	none = no text and OCR off
ocrConfidence	Tesseract mean word confidence 0–100	number	yes	90.2	null for text-layer pages
format	Text format	string	no	text, markdown
ocrLanguages	OCR languages used	array	yes	["eng"]	null when no page used OCR
charCount	Characters in text	integer	no	37820
text	The extracted text	string	yes	"W-9\n\nForm\n(Rev. March 2024)…"	null on error; "" when a page has no text
pages	Per page: pageNumber, method, charCount, ocrConfidence, text, error	array	no	see example	Document rows
title	Title from PDF metadata	string	yes	Form W-9 (Rev. March 2024)	null when the PDF has none
metadata	PDF metadata as in the file	object	yes	author, subject, creator, producer, creationDate, pdfVersion	Not verified. Document rows
encrypted	PDF is encrypted	boolean	yes	false	It may still be readable
error	Error of this row: code, message	object	yes	{"code": "NOT_FOUND", ...}	null when there is no error
warnings	Notes	array	no	["Only the first 50 of 120 pages were processed (maxPages)."]	Empty array when there is nothing to say
fetchedAt	When the file was downloaded (ISO 8601)	string	yes	2026-10-13T05:23:31.200Z	null when not downloaded
processedAt	When the row was written (ISO 8601)	string	no	2026-10-13T05:23:31.254Z

null means "not known or not applicable". 0 is a real count. A page with no text after OCR is complete with text: ""; it is a result, not an error. complete means the file was read; it does not mean the text is long. A scan read with ocr: never is complete with a warning: in document mode text is "" and each entry in pages has method: none; in page mode the single row has text: null and method: none.

Output entities and surfaces

Situation	Surface	Result shape	User action
Normal file	Dataset row	status: complete, text, pages	Use text
Some pages failed or the run stopped mid-document	Dataset row	Document mode: status: partial, pages[].error, warnings. Page mode: failed pages are rows with status: error; a stop is shown in warnings and RUN_SUMMARY	Check failed pages, run again for them
File not found (HTTP 404/410)	Dataset row	status: not_found, error.code: NOT_FOUND	Check the URL
Other problem	Dataset row	status: error, error.code, text: null	See Troubleshooting
Run totals	Key-value store RUN_SUMMARY	status, message, counts, errorCodes	Check chargedDocuments and ocrPageEvents

RUN_SUMMARY.status is one of SUCCEEDED, PARTIAL, ERROR, INVALID_INPUT, CHARGE_LIMIT_REACHED, DEADLINE_REACHED, TOO_MANY_FAILURES, CANCELLED. The run ends as FAILED only for invalid input or a platform error; a bad URL is an error row, and the run still succeeds.

Pricing

Pricing unit

This Actor uses pay-per-event pricing. You pay only for these events:

Event	When it is charged	Free	Bronze	Silver	Gold+
Actor start	Once per run, per GB of memory (default 1 GB)	$0.005	$0.005	$0.005	$0.005
document	A PDF read from the text layer (all its text-layer pages together), also when it has no text layer and OCR is off. In page mode: one per started 50 text-layer page rows.	$0.0025	$0.0024	$0.0023	$0.0022
ocr-page	One page or image read with OCR, once per selected language	$0.012	$0.0115	$0.011	$0.0105

Not charged: not_found and error rows, failed pages, pages skipped by pages or maxPages, and documents not processed because the run stopped. A run where every URL fails costs only the start event. Platform usage (compute) is included in the event prices.

The Actor checks your Maximum cost per run before every page. When the next page would go over it, the run stops, saves what it has, and marks the rest CHARGE_LIMIT_REACHED.

Cost examples (Free plan, 1 GB, English OCR)

Scenario	Input scope	Events	Cost
Prefill	1 text PDF + 1 image	1 document + 1 ocr-page	$0.0195
10 text PDFs	10 URLs	10 document	$0.030
100 text PDFs	100 URLs (one run)	100 document	$0.255
1,000 text PDFs	10 runs of 100 URLs	1,000 document	$2.55
10 scanned pages	1 scan, 10 pages	10 ocr-page	$0.125
100 scanned pages	1 run	100 ocr-page	$1.205
1,000 scanned pages	about 7–10 runs (see speed below)	1,000 ocr-page	about $12.04–12.05
10 scanned pages, 2 languages	languages: ["eng","deu"]	20 ocr-page	$0.245

Use cases

Buyer job	Input path	Output entity	Key fields	Next integration
Give an AI agent the text of a PDF link	urls	document row	text, pageCount, title	MCP tool call, LLM prompt
Split documents into pages for RAG	urls, outputMode: page	page row	pageNumber, text, method	Chunking, vector database
Read a scanned archive	urls, ocr: auto, languages	document row	ocrPages, pages[].ocrConfidence, text	Search index, review queue
Get headings and lists as Markdown	urls, outputFormat: markdown	document row	text, format	Notes, knowledge base
Read only part of long reports	urls, pages: "1-3"	document row	pagesProcessed, text	Summaries, alerts

Scheduling and monitoring

The Actor reads the URLs you give it at run time; it does not watch for changes. To read the same documents again on a timer (for example, a price list PDF that is updated every week), create an Apify schedule with a fixed input and compare text or charCount with the last run in your own system. Each run is charged again.

API, export and integrations

• API: start the Actor with POST https://api.apify.com/v2/acts/cliqtomedia~pdf-text-ocr/run-sync-get-dataset-items and the input JSON. You get the Dataset rows in the response.
• Export: download the Dataset as JSON, CSV, Excel, XML or HTML from the Output tab or the API.
• Webhooks: use Apify webhooks to send rows to your system when a run finishes.
• No-code: use the Apify integrations for Make, Zapier or n8n to pass text to the next step.

Use with AI agents (MCP)

• MCP server URL with only this Actor: https://mcp.apify.com?tools=cliqtomedia/pdf-text-ocr

• Smallest useful input:

  { "urls": ["https://www.irs.gov/pub/irs-pdf/fw9.pdf"] }
  

• Key output fields: status (did it work), text (the content), pageCount, ocrPages (how many pages needed OCR), error.code (what to fix).

• A text PDF takes a few seconds and costs $0.0075 with the start event. Each OCR page adds about 10 seconds and $0.012 per language.

• Limits: up to 100 URLs and 300 pages per document; public URLs only. For long scans, use pages or maxPages so the call returns in time.

Limitations and partial results

What it does not do

• No handwriting. Tesseract reads printed text; handwritten text gives poor or empty results.
• Tables are not rebuilt. Table text comes out as lines, without columns.
• No Chinese, Japanese or Korean OCR. PDFs in these languages still work if they have a text layer.
• No files behind a login, cookies or a private network. No uploads: only URLs.
• No DOCX, HTML or EPUB. An HTML page gives the error UNSUPPORTED_FILE_TYPE.
• A multi-page TIFF is read with one OCR call, which must finish within the time limit of one OCR page (50 seconds at 1 GB). Use PDF for long scans.

Speed and limits

• Up to 100 URLs per run. 100 text PDFs took 31 seconds in our test.
• Up to 300 pages per document (maxPages). A 300-page text PDF took about 17 seconds.
• OCR takes about 10 seconds per normal scanned page at 1 GB and up to 30–40 seconds for very dense small print. Plan for about 100–150 OCR pages per 30-minute run at 1 GB. At 2 GB OCR is about twice as fast for the same price per page.
• Memory: 1–2 GB (default 1 GB). Default run timeout: 1,800 seconds. The Actor stops 25 seconds before the timeout and saves what it has (DEADLINE_REACHED).
• File size: up to 100 MB (maxFileSizeMb, default 50). Images: up to 40 megapixels.
• Slow or hanging servers: one URL may take up to 45 seconds, one OCR page up to 50 seconds (half at 2 GB). If failed downloads, unreadable documents and failed pages waste more than 100 seconds in one run, the run stops and the rest are marked TOO_MANY_FAILURES, not charged.

Partial results

In document mode, a document gets status: partial when some pages were read and others failed (for example one OCR page timed out) or the run stopped in the middle of it (deadline or cost limit). Read pages keep their text; failed pages have pages[].error. Only read pages are charged. In page mode there is no partial status: each page row is complete or error, and a stop is listed in warnings and RUN_SUMMARY.

Troubleshooting

The row has UNSUPPORTED_FILE_TYPE

Symptom: status: error, error.code: UNSUPPORTED_FILE_TYPE. Likely reason: the URL opens an HTML page (a viewer or a download page), not the file itself. Check: open the URL in a private browser window. It should download or show the raw PDF or image. Action: use the direct file link.

The text is empty or strange

Symptom: charCount is low or the text has wrong characters; ocrConfidence is low. Likely reason: wrong OCR language, a blurry scan, handwriting, or a broken text layer. Check: pages[].method and pages[].ocrConfidence. Action: set the right languages. If method is text but the text is garbage, set ocr to always.

PDF_PASSWORD_REQUIRED or PDF_DAMAGED

The PDF needs a password or is broken. The Actor cannot open these files. Use a copy without a password.

DEADLINE_REACHED or TOO_MANY_FAILURES

DEADLINE_REACHED: the run was close to its timeout. Use a longer timeout, 2 GB memory, or fewer pages. TOO_MANY_FAILURES: too much time was lost on slow or failing URLs. Check those URLs and run the rest again. Unprocessed rows are not charged.

CHARGE_LIMIT_REACHED

The run reached your Maximum cost per run. Raise the limit or read fewer pages.

FAQ

Do I need a login or an API key for the documents?

No. The Actor reads public URLs only. It does not send cookies or passwords, and it does not open private network addresses.

Do you keep my documents?

No. Each file is downloaded to a temporary folder and deleted after it is read. Only the extracted text is saved to your Dataset. Make sure you have the right to process the documents you send.

Why did I pay for OCR on a PDF?

In auto mode, pages without usable text (scans, or pages with fewer than 25 visible characters or mostly broken glyphs) are read with OCR. Use ocr: never if you want the text layer only. With ocr: never a PDF is still charged as one document, also when it turns out to have no text layer.

Is OCR 100% correct?

No. Tesseract is good on clean printed text and weaker on low-quality scans, small print, handwriting and complex layouts. Check ocrConfidence.

Is the output the same as other PDF text extractors?

The text is similar, but field names are different (for example text and status). Map the fields when you switch.

Related Actors

• DOI Metadata & Bibliography Enrichment — get paper metadata by DOI, then read the open-access PDF here.
• YouTube Subtitles & Transcript Scraper API — text from videos for the same AI or RAG pipeline.

Support

For bugs, questions or feature requests, open an issue in the Issues tab of this Actor. Include the run ID, your input settings and the error.code. Do not send private documents, tokens or passwords.

OCR by Tesseract (Apache-2.0); PDF text by Poppler (GPL, run as a separate program).