pdf parsing

[fpfcmsr]

What problem does this solve?

searxng-mcp is great for gathering research about a topic, but lots of academic papers are only exposed as pdfs. Thus enabling searxng-mcp to read them would be very useful.

Proposed solution

a library like unpdf could be used to parse retrieved pdfs.

Alternatives considered

No response

Anything else?

No response

[ihor-sokoliuk]

Nice catch — this is a valid gap. The current URL reader explicitly classifies application/pdf as binary and returns an unsupported-content message, so academic papers published only as PDFs cannot currently be consumed through web_url_read.

I’ve recorded the work to add bounded PDF text-layer extraction using unpdf, including tests for size limits, malformed/encrypted files, and PDFs with no extractable text. Scanned/image-only PDFs will receive a clear unsupported/no-text result; OCR is intentionally out of scope for the first implementation.

I’ll keep this issue open as the action item.

[ihor-sokoliuk]

PDF parsing is available starting with v1.13.0.

Use web_url_read with the PDF URL as usual; no additional tool parameter is required. For application/pdf responses, the reader now extracts the text layer in a resource-bounded worker. Documents are limited to 500 pages, input and extracted text are capped at the lower of URL_READ_MAX_CONTENT_LENGTH_BYTES and 16 MiB, and parsing has a separate 30-second budget. Scanned/image-only and password-protected PDFs are reported clearly because OCR is not supported.

The behavior and limits are documented in the official URL Reader Controls documentation.

I’m leaving this issue open for now.