n8n-nodes-pdf2md

n8n community node that converts a PDF (binary) into semantic Markdown entirely in-process via @opendocsg/pdf2md. No external API, no network calls — the PDF never leaves your n8n instance.

Built after the n8n-nodes-pdf-extractor community node crashed (TypeError: …reading 'indexOf') on designed multi-page PDFs; @opendocsg/pdf2md parses them without crashing (heuristic font-size → heading detection).

Node: PDF to Markdown (pdf2md)

| Parameter | Default | Purpose | |---|---|---| | Input Binary Field | data | Binary property holding the PDF | | Output Field | markdown | JSON field the Markdown is written to | | Demote Headings One Level | false | Shift every heading down one level (H1→H2 … cap H6). Enable when the Markdown is nested under another H1 (e.g. PRIMA's # Content) so the document's own H1s don't end that section early. |

Output item = the input json plus { [outputField]: markdown }, binary preserved on the item. Honours Continue On Fail (emits { error } instead of throwing).

Build

npm install
npm run build      # -> dist/

Install into self-hosted n8n

Option A — publish to npm, then GUI install (easiest for users):

npm login
npm publish        # publishes n8n-nodes-pdf2md

Then in n8n: Settings → Community Nodes → Install → n8n-nodes-pdf2md.

Option B — install the packed tarball without publishing:

npm pack           # -> n8n-nodes-pdf2md-<ver>.tgz
# copy the tgz to the n8n host, then in the n8n custom-extensions dir:
cd ~/.n8n/custom && npm init -y 2>/dev/null; npm install /path/to/n8n-nodes-pdf2md-<ver>.tgz
# ensure N8N_CUSTOM_EXTENSIONS=~/.n8n/custom and restart n8n

Note

Heuristic extractors (this, unpdf, Jina-keyless) infer structure from font geometry — great on single-column reports/papers, noisier on heavily-designed magazine layouts. For pixel-perfect structure on designed PDFs, an LLM/vision converter (MarkItDown-with-LLM, Jina ReaderLM) is required.