PDF APIs you can pay per call

What a PDF API does

PDF is a page format: it records where glyphs, images and vector shapes sit on a page, and it can carry a text layer, form fields, metadata and attachments. A PDF API works on one of those layers. Some routes read the document and return text, Markdown or JSON. Others produce a PDF from a web page, an HTML string, Markdown or template data. A third group fills forms, merges files or sends a finished document onward.

Reading is not one operation. A digital PDF exported from a word processor already contains its text, and extraction recovers it with reading order, page numbers and metadata. A scanned PDF is a stack of page images, and recovering text from it requires optical character recognition, which is slower, depends on image quality and language, and can misread characters. Structured extraction goes a step further and maps the content to named fields, such as an invoice number, a date or a total.

Generation has its own choices. Rendering a live URL captures what the server returned when the call ran, including scripts that changed the page. Rendering your own HTML or Markdown gives you control over the layout. Template routes fill a fixed design with data, which suits invoices, reports and certificates that must look the same every time.

The output contract decides how you can use the result. Plain text is easy to index and loses tables. Markdown keeps headings and simple tables. Page JSON keeps the page boundary that a citation needs. A generated PDF may or may not be tagged for accessibility or suitable for long-term archiving. Read the response description before choosing on price.

Compare the returned data, sources and limits

The comparison separates six outputs that can qualify a service: text, Markdown or page JSON extracted from a PDF; OCR of scanned PDFs; invoice or schema fields extracted into JSON; PDF forms made fillable, filled or read back; PDFs generated from a URL, HTML, Markdown or template data; and existing PDFs merged into one file. Delivery by fax or for e-signature is listed as an adjacent product. It can accompany a qualifying service and never sets the row price.

A conversion route can accept more than PDFs. Document converters that also take web pages, DOCX or CSV appear here for their PDF path and keep their other inputs in the description. Image OCR routes read photos and screenshots rather than PDF pages, and they stay out of the comparison even when the same provider sells PDF extraction.

The OCR engine is named as Tesseract by Busker. A language or layout model is named in the extraction by agentsvc.io and agentutility. Model-based extraction handles irregular layouts and can also produce text that is not on the page, so review a sample before trusting it unattended.

Browser rendering is declared for PDF output by Agent Web Access and Scribe. A browser executes scripts and waits for content, which matters for single-page apps; it also means the result depends on timing, cookies and the server's response at that moment.

Tagged PDF/UA and PDF/A output is declared by makesPDF. European e-invoice XML formats are read by agentsvc.io, and Japanese documents are named as a strength by Mart402. A declared capability is the provider's claim; test it on your own documents.

Each price is the cheapest endpoint of that service for that kind of data: the same number wherever this page quotes it.

Compare 18 PDF services

Each row keeps the selected PDF endpoint and its reference price. Rows are ordered by observed paying wallets for the service; the order does not rank extraction accuracy, rendering fidelity or document limits.

$/CALL is the price of the endpoint named under it: the cheapest text, markdown or page json extracted from a pdf, ocr of scanned pdfs without a text layer, invoice or schema fields extracted from a pdf into json, pdf forms made fillable, filled or read back, pdf generated from a url, html, markdown or template data and existing pdfs merged into one file endpoint of the service or, for a service without one, its cheapest PDF endpoint. Never the cheapest tool of a gateway. Other kinds of data have their own price in the tables below. PDF ENDPOINTS counts the PDF endpoints out of all the endpoints the service lists. BUYERS 30D counts distinct wallets that paid the whole service on-chain in 30 days, not only its PDF endpoints. "Generalist" marks a gateway that sells more than three unrelated kinds of tools: it is never called the cheapest.

How to choose a PDF endpoint

Start from the document you actually have

Collect a handful of real files: a clean digital export, a scan, a document with tables, a long report and a form. Run each candidate on the same set. Price differences are small next to the cost of a route that returns empty text for scans or flattens every table.

Detect the text layer first

A cheap text-layer extraction answers most digital PDFs. When it returns little or no text, send the file to an OCR route instead of retrying the same extraction. A route may report that a file needs OCR or fall back to OCR on its own; check which behavior the provider describes.

Match the output to the next step

Search indexes need clean text with page numbers. Retrieval pipelines benefit from Markdown or page JSON that keeps headings and table structure. Accounting needs named fields. Pick the output format your next step reads, so you do not pay twice for a conversion and a reformatting.

Read the limits before the price

Page caps, file-size caps, character caps and time budgets differ by route. A route that stops at a fixed page count or returns a truncated result needs splitting logic on your side. Password-protected files and documents with unusual encodings can fail; test them explicitly.

Separate generation from extraction

If you need a PDF, the questions are layout control, page size, fonts, accessibility tags and whether the source is a live URL or your own markup. Rendering a URL depends on what the page serves; rendering your own HTML or Markdown is reproducible.

Test the operating limits

The monitored median uptime is 99.95% and the median selected quote is $0.01. Monitoring latency describes our own checks, not the processing time of a large document. Time a long file, a scan and a file with tables before committing a batch job to a provider.

Choose by the output your workflow needs

Start with the result your next step consumes. The candidates below use the same classification as the comparison table, and each job quote is the lowest priced endpoint of the requested types, with its path kept.

Read the text of a digital PDF

A digital PDF already carries its text. Extraction returns it as plain text, Markdown or page JSON; compare page numbering, reading order and table handling on a document you know before choosing a format for indexing.

Candidates today, most buyers first: agentsvc.io ($0.0055/call, generalist, 41 buyers) · DocEdge ($0.05/call, 39 buyers) · agentutility ($0.0025/call, generalist, 36 buyers) · WebberSites x402 Data API ($0.02/call, generalist, 33 buyers) · Tanod ($0.005/call, generalist, 26 buyers) · Mart402 ($0.01/call, generalist, 14 buyers) · Busker ($0.002/call, generalist, 13 buyers) · AgentWork Web Evidence ($0.01/call, 8 buyers) · url2md ($0.005/call, 6 buyers) · Scribe ($0.01/call, generalist, 6 buyers) · PageProof ($0.005/call, 3 buyers) · Kenward payable API ($0.02/call, 2 buyers) · x402.eleeth.com ($0.02/call, generalist, 1 buyer) · Basalt ($0.01/call)

Read a scanned PDF

A scan has no text layer, so a route has to recognize characters from the page image. Check the languages, page cap and confidence reporting, and keep low-confidence pages for review instead of indexing them silently.

Candidates today, most buyers first: agentutility ($0.0025/call, generalist, 36 buyers) · Mart402 ($0.01/call, generalist, 14 buyers) · Busker ($0.01/call, generalist, 13 buyers)

Extract invoice or schema fields

Field extraction turns a document into values your system can post. Keep the source file and the extracted value side by side, and check totals and tax lines arithmetically before an invoice reaches a ledger.

Candidates today, most buyers first: agentsvc.io ($0.02/call, generalist, 41 buyers) · Mart402 ($0.05/call, generalist, 14 buyers)

Fill or read a PDF form

A form route writes into or reads from named fields. Map your data to the form's own field names, and review a filled form before it is submitted on someone's behalf.

Candidates today, most buyers first: Emboss ($0.01/call, 2 buyers)

Generate a PDF from a page, HTML or data

Rendering a live page captures what the server returned at that moment, while a template route lays out your own data. Decide which one you need, and store the generation time with the file.

Candidates today, most buyers first: Agent Web Access ($0.005/call, 111 buyers) · agentsvc.io ($0.0045/call, generalist, 41 buyers) · Scribe ($0.02/call, generalist, 6 buyers) · makesPDF ($0.01/call, 1 buyer)

Merge existing PDFs

Merging changes the container, not the content. Confirm the page order, bookmarks and form fields of the result on a sample before merging documents someone will sign or file.

Candidates today, most buyers first: Relaystation ($0.01/call, 15 buyers)

What the providers describe

These paragraphs attribute output contracts, engines and limits to each provider. Reference prices follow the matching endpoint type, never an amount quoted in a marketing description.

Agent Web Access

Agent Web Access lists GET /pdf at $0.005. The provider says it renders a live web page to a PDF returned as base64, with JavaScript run first so single-page apps do not come out blank, and suggests it for archiving a page, a receipt or terms that may change. The service also reads pages as text and captures screenshots; those routes return web content rather than a PDF and stay outside this comparison. A rendered page is a snapshot of what the server delivered at that moment, so store the capture time with the file.

1 PDF endpoint of 4 · $0.005/call (GET /pdf) · Base · uptime 30d 99.6% · 209 ms · 111 buyers in 30 days

DocEdge

DocEdge lists POST /convert at $0.05. The service description says it converts a PDF, supplied by URL or as base64, into clean Markdown with extracted tables, metadata and an optional AI summary; the endpoint description adds webpages and Google Docs as inputs. Keep the summary apart from the extracted text: the first is generated, the second is the document's own wording.

1 PDF endpoint of 1 · $0.05/call (POST /convert) · Base/SOL · uptime 30d 99.9% · 1033 ms · 39 buyers in 30 days

agentutility

agentutility lists POST /pdf-to-markdown at $0.0025. The provider describes AI-powered, layout-aware extraction on the Datalab Marker engine for digital or scanned PDFs, preserving headings, tables, LaTeX equations, lists and multi-column flow, with Markdown, HTML or per-page JSON output and a 30-page maximum. A layout model can reorder or merge content on unusual pages; spot-check tables against the original before loading them into a database.

1 PDF endpoint of 9 · $0.0025/call (POST /pdf-to-markdown) · Base/SOL · uptime 30d 99.8% · 1502 ms · 36 buyers in 30 days

WebberSites x402 Data API

WebberSites lists GET /api/extract at $0.02: it fetches a PDF, DOCX or CSV by URL and returns Markdown plus structured JSON. For PDFs the provider describes text by page with metadata, and says the route flags scanned files that would need OCR. A flag tells you the text layer is missing; send flagged files to an OCR route instead of storing an empty result.

1 PDF endpoint of 37 · $0.02/call (GET /api/extract) · Base · uptime 30d 100% · 186 ms · 33 buyers in 30 days

agentsvc.io

agentsvc.io lists three PDF products. POST /pdf-extract, at $0.0055, returns the full text, per-page text with character counts, the page count and document metadata from a base64 PDF. POST /html-to-pdf, at $0.0045, renders a URL or an HTML string to PDF with page format, orientation and margins. POST /invoice-extract, at $0.02, turns a PDF, a photo or an XRechnung, ZUGFeRD or Factur-X file into invoice JSON. The provider says embedded e-invoice XML is read exactly without AI, while plain PDFs and photos are read by Claude. For an audit trail that difference matters: the XML is the invoice's own data, a model reading of a scan is an interpretation.

3 PDF endpoints of 50 · $0.0045/call (POST /html-to-pdf) · ARB/Base/POL/SOL · uptime 30d 100% · 56 ms · 41 buyers in 30 days

Busker

Busker separates text-layer extraction from OCR. POST /pdf, at $0.002, extracts Markdown from text-layer PDFs without OCR. Its OCR routes start at $0.01: POST /ocr runs Tesseract on scanned, machine-printed English documents up to 25 pages, and a deeper tier raises the cap to 100 pages under a time budget, returning the completed pages with a truncation reason when time runs out. The provider says PDFs that turn out to have a text layer are extracted the fast way at the same price.

3 PDF endpoints of 17 · $0.002/call (POST /pdf) · Base · uptime 30d 100% · 763 ms · 13 buyers in 30 days

Mart402

Mart402 lists GET /v1/parse at $0.01 for PDF to structured Markdown. The endpoint description quotes a flat fee per text PDF, a per-page fee for scanned OCR and a free quote route, so the figure shown here is a reference quote and the payment challenge for your file carries the amount. The provider describes its OCR as hallucination-guarded through dual-engine consensus and names Japanese as a strength. An invoice pipeline, POST /v1/document-pipeline at $0.05, adds field extraction, arithmetic verification and optional issuer enrichment.

2 PDF endpoints of 11 · $0.01/call (GET /v1/parse) · Base · uptime 30d 100% · 275 ms · 14 buyers in 30 days

PageProof

PageProof lists POST /v1/pdf/extract at $0.005: embedded text from a digital PDF into page-numbered JSON with SHA-256 hashes, up to 25 pages and 2 MiB. The provider is explicit that the route performs no OCR, summaries or URL fetching and does not verify claims made in the document. Store the returned hashes with the text so a later check can compare against them.

1 PDF endpoint of 1 · $0.005/call (POST /v1/pdf/extract) · Base · uptime 30d 99.7% · 2189 ms · 3 buyers in 30 days

Tanod

Tanod lists GET /v1/pdf at $0.005: text from a PDF at a public URL, with an optional page cap up to 200, returning the text, page count and document metadata. Its OCR route reads images rather than PDFs and stays outside this comparison. The provider describes its results as automated and heuristic, with a small free daily allowance per IP before payment.

1 PDF endpoint of 6 · $0.005/call (GET /v1/pdf) · ARB/Base/POL/SOL · uptime 30d 100% · 464 ms · 26 buyers in 30 days

Emboss

Emboss is a forms product. Its form routes start at $0.01: it turns a flat or scanned PDF form into a fillable one, fills a form from structured data or from supporting documents, and reads the values of a filled form back, with a completeness verdict when the form is known. Each endpoint description says the price scales with page count and that the payment challenge carries the amount for the PDF sent, so the figures here are reference quotes. A separate route faxes a PDF to a US, Canadian or Mexican number; that delivery service is listed as an adjacent product.

5 PDF endpoints of 5 · $0.01/call (GET /pay/read) · ARB/AVX/Base/POL · uptime 30d 99.9% · 171 ms · 2 buyers in 30 days

makesPDF

makesPDF lists GET /api/v1/markdown and GET /api/v1/render, starting at $0.01. The first converts GitHub-flavored Markdown to a tagged PDF; the second fills a template from structured data or an inline DSL. The provider says the output is dual-compliant with PDF/UA-1 and PDF/A-2A and is produced in about a second without a headless browser. Validate a sample with an accessibility and archiving checker before relying on the compliance claim in a regulated workflow.

2 PDF endpoints of 2 · $0.01/call (GET /api/v1/markdown) · Base · uptime 30d 100% · 1045 ms · 1 buyer in 30 days

Relaystation

Relaystation lists POST /v1/pdf/merge at $0.01 to combine PDFs in a chosen order. The endpoint description states a per-page rate with a minimum per x402 payment and says the remainder is credited to a balance. Check how that credit is applied before budgeting a batch of small merges.

1 PDF endpoint of 4 · $0.01/call (POST /v1/pdf/merge) · Base · uptime 30d 100% · 257 ms · 15 buyers in 30 days

Cost per document and for a larger batch

The median selected route costs $0.01 per call. At the same request shape, 100,000 calls total $1,000.00. The table repeats the calculation for the services named on this page; treat it as a reference figure for one document per call.

Each service at the price of its main PDF endpoint, the one named in the table above.

Document pricing often depends on the file. An endpoint description may quote per page, per document or by tier, or say that the payment challenge carries the amount for the file you send. A batch of long scans can cost far more than the reference quote suggests, while a batch of short digital PDFs can cost less. Request the challenge for a representative file before estimating a budget.

Two habits keep extraction costs down. Cache results by file hash, since a document that has not changed does not need a second extraction. Route by type, sending digital PDFs to text-layer extraction and only scans to OCR. Record whether a failed or empty extraction was charged under the provider's terms.

Inspect a request and its payment evidence

The public listing documents Agent Web Access at GET /pdf, which renders a live web page to a PDF returned as base64. This endpoint has a reference quote of $0.005. Pass the public URL you want to capture, then decode the base64 body to get the file.

The listed method and description establish the advertised contract. This explanation does not reproduce an HTTP 402 challenge or certify a paid response. Open the linked service's payment evidence and request the challenge for the exact call before authorizing a payment. Keep the resource, method, target URL, accepted payment and the returned file together, then open the PDF and compare it with the page in a browser.

How this comparison is built

The comparison starts with services that are online and payment-ready, have a known price, a description of at least 80 characters and measured 30-day uptime of at least 90%. Task selection then reads the documented endpoints. A service enters the table only with a priced route that extracts text or fields from a PDF, recognizes a scanned PDF, fills or reads a PDF form, generates a PDF or merges PDFs. Fax and e-signature delivery are shown as adjacent outputs. Image OCR, web-page readers, a PDF receipt of a payment and services whose captured endpoints include no PDF route stay out, even when the service description mentions PDF. Pages below the index threshold can still be read directly.

Provider claims were reviewed against the public service and endpoint descriptions captured on October 22, 2026, and remain provider claims. Monitoring supplies operating figures and reference prices; it does not test extraction accuracy, OCR quality or rendering fidelity. Buyer counts describe observed paying wallets for the service, with the coverage limits explained on its linked page.

The visible comparison date is October 26, 2026. Endpoint membership, row prices, the type table and the job candidates share one classification, so a cheaper delivery route cannot replace the quote of a PDF output.

Questions

How to pay

Each service answers an unpaid call with HTTP 402 and its price; your agent signs a USDC payment on the listed network and retries. No API key, no account, no subscription. See what x402 is.