Hey n8n Community,
Just wrapped a build that takes incoming documents from one team, bundles them with assets from a second team, and routes the package to two sequential signers, all triggered by emails. Some learnings that might save you a few hours.
1. Order your state machine for the human, not the system
Originally I wired: collect input from team A → send for signature → ask team B for supporting docs → loop back for re-sign. Total mess. Swapped to: collect from team A → wait for team B → bundle → single signature round. Rule I’m keeping: signers should never have to context-switch back to a document they thought was done. Re-route the state machine if you have to.
2. Filesystem-mode binary breaks $binary.data.toString('base64')
If your n8n runs in filesystem storage mode (the default for self-hosted), binary.data is just a marker string, not the actual file bytes. Calling .toString('base64') returns base64 of the marker, which is ~14 bytes of garbage that downstream APIs happily accept then mangle. Use this.helpers.getBinaryDataBuffer(0, 'data') inside a Code node to load real bytes.
3. The Merge trick for cross-node binary access
After a Sheets Lookup or Update, your binary is gone from the current item. Instead of $('NodeName').item.binary.data references everywhere, branch your binary-source node into two paths: one through your normal chain, one to a Merge node (Combine by Position) downstream. The Merge stitches the binary back exactly where you need it. Way cleaner than cross-node refs.
4. Parser nodes should be lenient on attachment types, strict on validation
I started by requiring a PDF attachment. Then real-world users sent two phone photos and zero PDFs. Then a PDF plus inline-embedded images from email signatures. Then nothing at all by accident. Build your parser to classify by mime type AND filename, accept what’s plausible, and only throw when there’s literally nothing to process. The downstream “should I run extraction?” decision belongs in an IF node, not in the parser.
5. n8n’s JSON body validator runs before expressions resolve
If your HTTP node body has a JSON expression like "documents": {{ JSON.stringify($json.documents) }}, n8n’s parser flags it as invalid JSON before the workflow runs, because at validation time it sees {{ ... }} as literal text, not parsed JSON. Fix: wrap the entire body in one big expression, prefixed with =:
={{ JSON.stringify({ template_id: ..., documents: $json.documents }) }}
Single expression, evaluates at runtime, validator stays happy.
Bonus: extraction from variable-quality input
For pulling structured data out of incoming attachments (could be a clean scan, could be a handwritten form photographed on a phone), I used my own Extractor. Way less brittle than OCR + regex.
-
n8n Cloud: search
easybitsin the nodes panel, it’s verified by n8n, so it shows up natively, no install step -
Self-hosted: install the community node
@easybits/n8n-nodes-extractorfrom Settings → Community Nodes
What’s the messiest real-world input your workflow has had to handle? Curious whether others are dealing with the same “user sends whatever, parser figures it out” situation, or if you’ve found a clean way to enforce structure upstream.
Best,
Felix