PDFs in.
Markdown out.

Try it without a key

Put our URL in front of any PDF link to get Markdown back. It works in a browser, in curl, or in any agent that can fetch a URL.

Without a key you get one test call a day, up to 10 text pages, without OCR. For more, buy credits.

What you get back

Markdown you can put straight into a prompt, plus details about the file so your app can decide what to do next.

Set it up with one prompt

The prompt points your AI builder at our machine-readable docs, so it writes the integration correctly, including keeping your key on the server.

Add PDF upload to my app using PDF Inspector.
Read https://pdfinspector.com/llms.txt first and follow it.

- Let users upload a PDF (max 50 MB).
- Send it to POST https://api.pdfinspector.com/v1/convert
- Use the API key from the secret PDFINSPECTOR_API_KEY.
  Call the API from a server function, never from the browser.
- Show a progress state (scanned PDFs can take a few seconds).
- Show the returned "markdown" nicely formatted.
- Then send the markdown to our AI and summarize the document.

claude mcp add --transport http pdfinspector https://pdfinspector.com/mcp

Convert ./q3-report.pdf to Markdown and pull the revenue table into a CSV.

{
  "mcpServers": {
    "pdfinspector": {
      "url": "https://pdfinspector.com/mcp",
      "headers": { "Authorization": "Bearer sk_live_YOUR_KEY" }
    }
  }
}

Fastest in the benchmark, and best at tables

pdf-inspector's own benchmark on the 200-PDF opendataloader-bench corpus, comparing local parsers without OCR.

Source: pdf-inspector README, refreshed July 31, 2026 on an Apple M4 Pro. Native-text extraction only. The hosted API adds network time, and OCR pages take a few seconds.

We don't keep your files

People upload contracts, bank statements, medical letters and CVs. PDF Inspector holds each file only while converting it, then deletes it. This applies on every plan.

Deleted after processing

Files exist only while your request runs.

Not logged

File contents and extracted text never appear in our logs, error logs included.

Not used for training

Your documents never train AI models.

Results expire

Long-document results are deleted when you fetch them, or after one hour.

What we keep: timestamp, key ID and page count, for billing. Details

How an agent uses PDF Inspector

Every step is an HTTP call or an MCP tool, and every error says what to call next. Paying is the one step that needs a person.

1. Test without a key

   One call a day, up to 10 text pages.

   GET https://api.pdfinspector.com/md/https://example.com/report.pdf
   
   200 OK · text/markdown
   # Q3 Report …

2. Create a key

   The key is returned right away, with no credits. The agent gives its user the checkout link.

   POST https://api.pdfinspector.com/v1/keys
   { "email": "you@example.com" }
   
   201 Created
   { "key": "sk_live_…", "credits_remaining": 0,
     "checkout_url": "https://pdfinspector.com/checkout/…" }

3. Run out of credits

   The error names the next step and links to checkout for the user.

   402 Payment Required
   { "error": { "code": "insufficient_credits",
     "next_action": { "type": "buy_credits", "url": "https://pdfinspector.com/checkout/…" } } }

Pricing

Text pages cost 1 credit each, and scanned pages that need OCR cost 15. Buy credit packs as you need them, or subscribe for a lower price per page.