Description

Free PDF to Text for n8n


Extract text from PDF files inside your self-hosted n8n instance. No API key, no per-page fees, and your documents never leave your server.

Features

• Local processing: PDFs are parsed in memory by PDF.js. The parser receives no URL and makes no network requests for your document.
• Readable output: line breaks and table columns are preserved, which helps regular expressions and AI nodes.
• Flexible output: one item per PDF or one item per page, with an optional page range.
• Password support: open protected PDFs, with clear errors for a missing or incorrect password.
• Scan detection: a needsOcr flag tells a workflow when a PDF contains no text layer.
• Keeps n8n responsive: parsing runs in a worker thread, with enforced limits on time and memory.
• Tolerant of damaged files: malformed cross-reference tables are recovered where possible, and one unreadable page does not stop the others.

Installation

1. In n8n, go to Settings > Community Nodes.
2. Select Install a community node.
3. Enter n8n-nodes-free-pdf-to-text and confirm.
4. Search the node picker for Free PDF to Text.

This is an unverified community node for self-hosted n8n, because it needs PDF.js as a runtime dependency.

n8n 3

n8n 3 refuses to install unverified community nodes by default. Set this environment variable on your n8n instance and restart it before installing:

N8NUNVERIFIEDPACKAGES_ENABLED=true

Without it, the installation fails with Package n8n-nodes-free-pdf-to-text is not vetted for installation.

Set the variable before you upgrade from n8n 2 as well. n8n’s breaking changes state that installed unverified nodes stop loading without it.

Quick start

1. Add a node that outputs a PDF as binary data, such as HTTP Request, Webhook, or Read/Write Files from Disk.
2. Connect Free PDF to Text and set Binary Property to the field that holds the PDF (data by default).
3. Run the workflow and use pages, text, or needsOcr in the next node.

Output

One item per PDF (default)

{
	"filename": "invoice.pdf",
	"totalPages": 3,
	"pages": ["Invoice 2026-041nDate: 4 November", "Totalt100,00nVATt21,00", ""],
	"pageNumbers": [1, 2, 3],
	"characterCount": 54,
	"needsOcr": false,
	"pagesWithoutText": [3],
	"warnings": [{ "page": 3, "message": "Recovered malformed content stream" }],
	"text": "All pages as one value, when Include Raw Text is on",
	"metadata": {},
	"info": {}
}

One item per page

{
	"filename": "invoice.pdf",
	"page": 2,
	"totalPages": 3,
	"text": "Totalt100,00nVATt21,00",
	"characterCount": 22,
	"hasText": true
}

Field reference

| Field | Meaning |
| —————— | ——————————————————————————- |
| totalPages | Pages in the whole document, also when a page range is set |
| pages | Text of the selected pages, in page order |
| pageNumbers | PDF page number of each entry in pages |
| characterCount | Number of extracted characters |
| needsOcr | true when no selected page contains text |
| pagesWithoutText | Selected pages that were read successfully but contain no text |
| warnings | Pages that could not be read. Those pages are empty and the others still return |
| text | All pages joined by a blank line, when Include Raw Text is on |
| metadata, info | Document metadata, when Include Metadata is on |

Within a page, a newline separates lines and a tab separates widely spaced columns.

Options

| Option | Default | Purpose |
| —————————- | ———-: | ——————————————————- |
| Binary Property | data | Binary field containing the PDF |
| Include Raw Text | Off | Add all text as one value in text |
| Include Metadata | On | Add PDF metadata and document information |
| Output Mode | One per PDF | Return one item per PDF or one item per page |
| Page Range | All pages | Pages to extract, such as 1-3, 5, 8- |
| Password | Empty | Password for a protected PDF |
| Keep Input Binary | On | Pass the PDF on to the next node |
| Maximum File Size | 50 MB | Reject larger PDFs before parsing (up to 100 MB) |
| Maximum Pages | 500 | Reject when more pages would be extracted (up to 2,000) |
| Maximum Extracted Characters | 5,000,000 | Stop on excessive text output (up to 10,000,000) |
| Maximum Memory | 1,024 MB | Stop parsing above this memory use (128 to 8,192 MB) |
| Timeout | 120 seconds | Stop processing of one PDF (up to 300 seconds) |
| Strict Parsing | Off | Fail on malformed PDFs instead of attempting recovery |

Scanned PDFs

The node reads existing text and does not perform OCR. A scan with a hidden OCR layer works as usual. A scan without one returns empty pages and needsOcr: true, so you can route it with an If node:

Free PDF to Text → If {{ $json.needsOcr }} → true:  your OCR step
                                           → false: continue with the text

needsOcr only looks at whether text exists. Use pagesWithoutText to find individual scanned pages in a mixed document, keeping in mind that a page in that list can also be intentionally blank.

Node versions

Version 2 is used for every new node. It keeps line breaks and table columns.

Version 1 nodes in existing workflows keep their space-separated text, so an update does not break expressions that depend on it. To switch, replace the node with a new one and check any expressions or regular expressions that read its text.

Security

• PDF JavaScript, XFA, worker fetching, WASM, image decoding, and font rendering are disabled.
• Parsing runs in a worker thread that is terminated at the timeout or memory limit, and when the execution is cancelled.
• File, page, and character limits reduce the impact of malformed or hostile documents.
• The Password option is stored in the workflow like any other node parameter. To avoid that, set it with an expression that reads the password from an earlier node or an environment variable.

Run any document parser that handles untrusted files in an isolated and up-to-date environment.

Compatibility

• Self-hosted n8n 2.x and 3.x. CI tests the current stable release and n8n 3.1.0, which is an early access release at the time of writing.
• n8n 3 requires N8NUNVERIFIEDPACKAGES_ENABLED=true, see Installation.
• Node.js 22.22 or newer.
• No credentials or external accounts.

Development

pnpm install --frozen-lockfile
pnpm lint
pnpm test

pnpm test builds the package and also runs the worker-thread tests against the compiled output. Two more tests use a real n8n container and require Docker:

• pnpm test:n8n [tag] runs the packed local package through a workflow.
• pnpm test:n8n-install [tag] installs the version published on npm through n8n itself.

See CONTRIBUTING.md for the full workflow, and CHANGELOG.md for release history.

Automatic updates

On the first of every month, a GitHub Actions workflow checks pdfjs-dist, @n8n/node-cli, and prettier. It only considers stable versions within the current major version that have been public for at least seven days.

• New pdfjs-dist: after all tests pass, it is released to npm automatically as a patch version.
• Tooling only: the update is committed without a release, because users would install nothing new.
• Failure: nothing is pushed or released, and an issue is opened.

Support

Report bugs and request features in GitHub Issues. Please include your n8n version, the node version, relevant options, and a non-sensitive sample PDF if possible.

Credits and license

Maintained by LeonardSEO. Based on the MIT-licensed n8n PDF Page Extract node by Oscar Weijman, who created the original page-by-page extraction. This fork adds a modern parser and the features above.

Released under the MIT License.