micropdfBETA Start building
BUILT FOR AGENTS. OPEN TO BUILDERS.

PDFs in.
Context out.

Give your agent documents it can actually read. Turn PDFs into Markdown and structured JSON, with a price for every page.

Read the quickstart See the output

✓ No account to create ✓ Pay with x402

Public beta on Base Sepolia testnet
DOCUMENT → CONTEXT01 / EXTRACT
↳field-notes.pdf2 pages · sample documentPDF
A little less paperwork.
Words. Tables. Something useful.
μextract
ready for your agentJSON 
{
  "page": 1,
  "method": "native",
  "blocks": […]
}
Illustrative output. Real source references.
ONE SMALL API. LESS DOCUMENT FRICTION.Markdown + JSONNative text + OCRx402 paymentsPage provenance

THE USEFUL PARTS, EXTRACTED.

From document to
the next step.

Your agent needs the content, where it came from, and a clear way to get it.

01 / READ

Text, even in scans.

Extract native PDF text. Use English OCR when a page has little embedded text, including mixed native and scanned documents.

Native extraction + printed English OCR
02 / STRUCTURE

A format that fits.

Get readable Markdown or paginated JSON with text blocks, headings, lists, and conservatively detected simple tables.

One extraction, two output formats
03 / TRACE

Keep the source close.

Page numbers and block coordinates help your agent point back to the source. Partial results and warnings stay explicit.

Page references + layout warnings

LESS GUESSWORK. MORE CONTEXT.

Readable by humans.
Usable by agents.

Feed Markdown into your workflow, or use JSON when page boundaries and source coordinates matter.

Explore the response format

Example response. Layout and tables are heuristic; always inspect warnings.

SAMPLE OUTPUT 
<!-- page: 1 -->

# Field notes

A small study of useful things.

| Item | Count |
| --- | --- |
| Good questions | 24 |
| New ideas | 12 |

<!-- page: 2 -->

## What happens next

Turn these notes into something useful.
Same job. Your choice of format.

HTTP ALL THE WAY DOWN.

A workflow your agent
can follow.

Read the integration guide ↗
  1. 01

    Send a PDF.

    Upload a file or provide a public HTTPS URL. Get an inspection and a document token.

    POST /v1/documents
  2. 02

    Quote. Then pay.

    Select pages, review the price, and authorize the x402 payment with your agent’s wallet.

    402 → PAYMENT-SIGNATURE
  3. 03

    Pick up the context.

    Poll the job and fetch Markdown or JSON. Delete the document when you’re done.

    GET /v1/jobs/{id}/result

MICRO PRICE. NO MONTHLY COMMITMENT.

Pay for pages.
Keep building.

One quoted price per selected page. Native text and OCR use the same rate. No subscription or prepaid credit balance.

How x402 payments work ↗
EXTRACTIONTESTNET BETA
0.001 USDC / page

Current Base Sepolia testnet rate.

10 pages 0.010 USDC50 pages 0.050 USDC
  • Markdown and JSON included
  • Up to 20 MB and 50 pages per PDF
  • Source and results expire after 24 hours
Make your first request ↗The quote is authoritative. Mainnet payments are not enabled.

BEFORE YOU BUILD.

A few good
questions.

Do I need an account or API key?

No account is required. Uploading returns a secret document token that authorizes access to its quotes, jobs and results. A compatible wallet is required for x402 payments.

Is this ready for real-money payments?

Not yet. This public beta uses test USDC on Base Sepolia. A real signed testnet settlement still needs end-to-end validation. Use the beta to evaluate integration and extraction quality.

What kinds of PDFs work?

Native-text PDFs and scans of printed English. The current limits are 20 MB and 50 pages. Handwriting, complex tables, formulas, and image descriptions are outside the current scope. Encrypted PDFs are rejected.

What happens to my documents?

Source PDFs and results expire 24 hours after upload. You can delete them earlier through the API. Minimal payment and job audit records remain; unresolved payment records remain until reconciled. Don’t use the beta for sensitive documents without assessing these limits.

What if extraction or payment fails?

Check job status, partial results, and warnings. Reuse the same quote and idempotency key for retries. If settlement is uncertain, stop and retain the quote ID for operator reconciliation. Automatic refunds are not implemented.

ONE LESS THING BETWEEN YOUR AGENT AND THE ANSWER.

Let the documents in.

Start with the docs ↗