PDF Operations Toolkit — Merge, Split, Compress, Rotate, Protect, Unlock, Watermark, Extract Text & More

Thirteen PDF operations in one pay-per-operation API: merge, split, reorder, extract pages, delete pages, rotate, compress, rasterize (page images), protect (AES-256 password + print/copy restrictions), unlock (your own password), watermark, info (page count, page sizes, outline, metadata, encryption, text-layer presence) and extract text (per page, optional tables). Input PDFs come in as public URLs, base64 / data: URLs or keys in the run's key-value store; output files land in the run's key-value store with a SHA-256 each. Every operation returns a manifest that reconciles every input page to an output page — the actor never loses a page silently, and deleted pages are counted in pages_dropped. Pay only for operations that completed and reconciled.

• Chain operations in one run: a list of {op, inputs, params} objects; each gets its own manifest row and output keys (op-0-…, op-1-…).
• Password-protected files: password at the top level or params.password per operation opens encrypted inputs; the password is never guessed and never written to the output. protect sets a new user/owner password with optional disallowPrint / disallowCopy.
• Page selectors: params.pages as "1-3, 7, 9-" or a list of page numbers for extract_pages, delete_pages, rotate and extract_text; params.ranges [[from, to], …] for split; params.order for reorder.
• Safe by construction: byte and page ceilings, decompression-bomb guard, corrupted inputs rejected loudly, embedded JavaScript / auto-actions only reported (active_content_flags), URLs fetched from public hosts only with robots.txt honoured and an identified User-Agent.

Billing: $0.02 per run start, then $0.05 per completed, reconciled operation (pdf_operation_check). Rejected inputs (oversized, corrupted, wrong password, not a PDF, unreachable URL) and non-reconciling operations are free status rows.

Who uses this

• Developers automating document pipelines — merge statements, split scanned batches, strip cover pages, stamp drafts — without running a PDF server.
• Legal, finance and operations teams who need protected, watermarked or compressed deliverables with a page-by-page audit trail.
• Data and AI teams pulling per-page text and tables from PDFs by URL as one step in a larger workflow.

📊 Sample Output

Real rows from run 1j15vJcRwAvSUMa2U on build 0.1.27 (2026-10-19), the Quick start input below (long values shortened):

op	outcome	inputs	input_sources	input pages	output pages	pages_dropped	output_keys	reconciled	results (start)
split	answer	1	["url"]	1	1	0	["op-0-split_1_p1-1.pdf"]	True
info	answer	1	["url"]	1	0	0	[]	True	[{"input_ref":"sample.pdf","page_count":1,"metadat…

The run finished with the status message: 2 operation(s): 2 reconciled+charged, 0 unbilled (rejected/unparseable) | pdf_operation_check charged=2

✅ What you get

One pdf_operation_manifest row per operation (outcome: answer rows are billed; rejected / unparseable rows are free):

• op, input_count, input_refs, input_sources (url / base64 / kvs per input), input_page_total, input_sha256s.
• output_count, output_keys (files in the run key-value store), output_page_total, output_sha256s, size_delta_bytes.
• page_map — every output page traced to its input file and page; pages_reconciled (true = no page lost), pages_dropped (delete_pages).
• results — the payload of info (page_count, page_sizes, toc, metadata, encrypted, has_text) and extract_text (per-page text, optional tables).
• active_content_flags (JavaScript, OpenAction, launch actions, embedded files — reported, never run), warnings, faults, outcome_reason.

⚙️ Sample inputs

1. Quick start — split the first page and inspect a public PDF (the Store example)

{
  "operations": [
    {"op": "split", "inputs": ["https://pdfobject.com/pdf/sample.pdf"], "params": {"ranges": [[1, 1]]}},
    {"op": "info", "inputs": ["https://pdfobject.com/pdf/sample.pdf"]}
  ]
}

About $0.12 on the Free tier ($0.02 start + 2 × $0.05).

2. Merge three files, delete the cover page, watermark and protect the result

{
  "operations": [
    {"op": "merge", "inputs": ["https://example.com/a.pdf", "https://example.com/b.pdf", "https://example.com/c.pdf"]},
    {"op": "delete_pages", "inputs": ["https://example.com/a.pdf"], "params": {"pages": "1"}},
    {"op": "watermark", "inputs": ["https://example.com/a.pdf"], "params": {"text": "CONFIDENTIAL", "opacity": 0.2}},
    {"op": "protect", "inputs": ["https://example.com/a.pdf"], "params": {"userPassword": "open-me", "ownerPassword": "owner-only", "disallowPrint": true}}
  ]
}

3. Unlock your own protected file, rotate two pages and extract text with tables

{
  "operations": [
    {"op": "unlock", "inputs": ["my-upload-key"], "params": {"password": "open-me"}},
    {"op": "rotate", "inputs": ["my-upload-key"], "params": {"degrees": 90, "pages": [2, 3], "password": "open-me"}},
    {"op": "extract_text", "inputs": ["my-upload-key"], "params": {"pages": "1-5", "tables": true, "password": "open-me"}}
  ]
}

my-upload-key is a PDF you stored in the run's key-value store; a base64 string or data:application/pdf;base64,… works the same way.

🧾 JSON sample record

{
  "op": "split",
  "input_count": 1,
  "input_refs": "[\"sample.pdf\"]",
  "input_page_total": 1,
  "input_sha256s": "[\"229defbb0cee6f02673a5cde290d0673e75a0dc31cec43989c8ab2a4eca7e1bb\"]",
  "output_count": 1,
  "output_keys": "[\"op-0-split_1_p1-1.pdf\"]",
  "output_page_total": 1,
  "output_sha256s": "[\"c157c3b76cd8a04e8423d3c60c21c7e1dac0d25036ebca2824081dc79eeac8e1\"]",
  "page_map": "[{\"output\":\"split_1_p1-1.pdf\",\"output_page\":1,\"input_ref\":\"sample.pdf\",\"input_page\":1}]",
  "pages_reconciled": true,
  "size_delta_bytes": -1200,
  "active_content_flags": null,
  "warnings": null,
  "faults": null,
  "pages_dropped": 0,
  "results": null,
  "input_sources": "[\"url\"]",
  "record_type": "pdf_operation_manifest",
  "source": "buyer-supplied PDFs (URLs / base64 / data: URL / key-value-store keys)",
  "outcome": "answer",
  "input": "split(1 input)",
  "outcome_reason": null
}

🔧 How it works

Resolve. Each input is fetched once: URLs through the public-host (SSRF) guard and the site's robots.txt, with redirects walked hop by hop; base64 / data: URLs decoded; key-value-store keys read. Anything that is not a PDF, is over maxBytes / maxPages, or is encrypted without a working password is rejected as a free row.

Operate. PyMuPDF performs the operation in memory. Encrypted inputs are authenticated with the password you gave. protect, unlock and compress rebuild the document into a fresh file so damaged cross-reference tables are repaired on the way out.

Reconcile. The page map traces every output page to its input file and page number; pages_reconciled is true only when input pages = output pages + pages_dropped. An operation that would lose a page is delivered unbilled with the reason.

Billing. Each reconciled manifest is pushed and charged in one call; a row that cannot be charged is not delivered. A run whose every URL input was blocked or unreachable ends SUCCEEDED with SOURCE UNAVAILABLE … nothing charged.

Limits. 50 MB and 1,000 pages per input by default (maxBytes up to 200 MB, maxPages up to 5,000); rasterize 36–400 DPI; output files are kept in the run's key-value store for its retention period.

💰 Pricing example

Event	Free	Bronze	Silver	Gold
Actor Start (apify-actor-start)	$0.02	$0.02	$0.02	$0.02
PDF operation (pdf_operation_check)	$0.05	$0.045	$0.04	$0.0335

• 10 operations: $0.02 + 10 × $0.05 = $0.52
• 1,000 operations: $0.02 + 1,000 × $0.05 = $50.02

A run that delivers zero reconciled operations charges the $0.02 start fee only. The start fee is charged once per GB of run memory; the default run memory is 1024 MB.

⚖️ Legal & ToS

Operates only on PDFs you supply or point to by public URL; URL fetches honour robots.txt, use an identified User-Agent and never log in. No content inside a PDF is executed. Your inputs and outputs stay in your own run storage. Use unlock only on files you are entitled to open.

❓ FAQ

Q: Where are the output files? A: In the run's key-value store under the keys listed in output_keys (Storage → Key-value store, or the API). Download them within the store's retention period.

Q: Can one operation use the output of another? A: Not inside a single run yet — operations read the inputs you give. Run twice (second run's inputs = first run's output keys copied to a named store) or do the chain as separate operations on the same source as in sample 2.

Q: Why was my operation rejected as "encrypted / password-protected"? A: The file needs a password and none (or the wrong one) was given. Add password at the top level or params.password; the actor never guesses.

Q: What does pages_reconciled: false mean? A: The output has fewer pages than the input minus the pages you asked to drop — the operation is delivered unbilled with the reason so nothing is lost silently.

🆘 Troubleshooting

• input resolution failed: blocked → the URL's host is private, its robots.txt forbids the path, or the server answered 403/429/5xx. Free. Try a public direct link.
• not a PDF (content-type …) → the URL served an HTML page (often a viewer or login page), not the file; use the direct download link.
• over the byte / page ceiling → raise maxBytes / maxPages within the limits.
• Run TIMED-OUT → split the operations across runs or raise the run timeout; delivered rows are kept.

🔗 Related Actors

• PDF Text Extractor — Markdown, per-page text, tables, OCR and RAG chunks with a readiness verdict.
• HTML to PDF Evidence Renderer — turn web pages into hashed PDF evidence.
• 🏢 About NexGen Watch — more than 300 public-data actors, pay-per-result: apify.com/nexgenwatch.

⭐ Found this useful?

A quick review on the Apify Store helps other teams find it. Feature request? Open it from the Issues tab — every one is read.