PDF Fetch
On this page
Purpose
PDF Fetch downloads public HTTP or HTTPS PDFs and extracts text from each page. It returns either one combined Text message or a Pages table with one row per page.
It extracts text-based PDFs with pypdf. It does not perform OCR, so scanned PDFs can return little or no text.
When to use it
- Read a public research paper or report.
- Extract text from clinical guidance, product information, or a committee paper.
- Keep page markers so a model can cite the source page.
- Filter page-level rows before later processing.
Required setup
Have at least one direct public HTTP or HTTPS PDF URL. The URL must work without cookies, authorization headers, or access to a private network.
How to use it
In Tool Mode:
- Connect the component tool to Agent Core.
- Keep the timeout, page, byte, character, and page-marker limits set on the component.
- Give the Agent a direct public PDF URL at runtime.
- Use the returned Text in the Agent response. Pages remains a Normal-mode output.
For example, give the Agent a public clinical guideline PDF and ask it to summarize a named section with page markers.
In Normal mode:
- Add one or more direct public PDF URLs.
- Keep Include Page Markers enabled when page citations matter.
- Connect Text to Split Text or Chat Output.
- Connect Pages to DataFrame Operations or Parser when you need page-level filtering or formatting.
Each URL runs independently. A bad URL reports an error for that URL without stopping the others.
Configuration
| Setting | What it controls | Recommended starting point |
|---|---|---|
| URLs | Public PDF links to fetch | Start with one direct PDF URL |
| Timeout | The request limit for each URL | Keep the 60-second default |
| Max Pages | The first N pages to extract | Use a positive limit for large documents |
| Max Bytes | The largest response accepted per URL | Keep the 20 MiB default |
| Max Characters per URL | The Text-output cap for each URL | Keep 100,000 for Agent or model context |
| Include Page Markers | Adds [page N] to Text output | Leave enabled for source-aware answers |
0 Max Pages means all pages. 0 Max Characters per URL disables Text truncation. The character cap does not affect Pages output.
Expected result
Text combines each PDF under its URL header. Pages returns url, page, text, and error columns. Empty input produces empty output.
PDF Fetch accepts only public HTTP(S) addresses. It rejects unsupported schemes and private, loopback, link-local, reserved, and multicast addresses. It cannot use cookies, authorization headers, or private document stores.
Reference details
Lower Timeout, Max Pages, or Max Characters when the next component has a small context window. Use Pages when you need the full extracted page text because the Text cap does not apply there.
Split Text can chunk extracted Text or Pages. Use DataFrame Operations to filter Pages rows, Parser to format them into a Message, or Chat Output to display extracted content.