n8n-code-python
Write Python in n8n Code nodes (native Python, `language` pythonNative, n8n 2.x). Use when the user explicitly wants Python in a Code node, when migrating old Pyodide/"Python (Beta)" code that used _input/_json/_node/_now, or when a Python Code node fails with NameError, "Security violations detected", "Import of standard library module … is disallowed", "__build_class__ not found", "A 'json' property isn't a dictionary", or "Python runner unavailable". Covers the only two variables (_items/_item), dict-only access, imports blocked by default, the sandbox's denied builtins, accepted return shapes, and how errors interact with onError. JavaScript is the default for Code nodes — native Python has no n8n helpers and, by default, no imports. EXCEPTION — for Python in the AI-agent-callable Custom Code Tool (@n8n/n8n-nodes-langchain.toolCode), use the n8n-code-tool skill instead (input is _query, return must be a string).
Resumen
Python Code Node (native)
Since n8n 2.0 the Code node's Python runs as native Python in a task runner (language: "pythonNative"). The old Pyodide "Python (Beta)" is gone, and with it every n8n helper it had: _input, _json, _node, _now, _today and _jmespath all raise NameError now. Code copied from old templates, forum posts or older docs usually breaks on this.
JavaScript first — stronger than before
Native Python gives you two variables and plain Python, and by default no imports: no json, datetime or re. Everything n8n-specific ($('Node'), $jmespath, Luxon, this.helpers.httpRequest, static data) exists only in JavaScript. Use Python only when the user explicitly asks for it. Even then, first check whether an expression, Edit Fields, or a native node (Crypto, Date & Time, HTML, XML) does the job. See n8n-code-javascript and the transform gatekeeper in n8n-expression-syntax.
Quick start
# Run Once for All Items (default mode)
return [
{"json": {"name": it["json"]["name"], "revenue": it["json"]["revenue"]}}
for it in _items
if it["json"].get("active")
]# Run Once for Each Item
row = _item["json"]
return {"json": {**row, "name_upper": (row.get("name") or "").upper()}}Node config: {"language": "pythonNative", "mode": "runOnceForAllItems" | "runOnceForEachItem", "pythonCode": "..."} on n8n-nodes-base.code typeVersion 2.
The only inputs: _items and _item
| Mode | Variable | Shape |
|---|---|---|
| Run Once for All Items | _items | list of plain dicts {"json": {...}, "pairedItem": {...}} |
| Run Once for Each Item | _item | one plain dict {"json": {...}, "pairedItem": {...}} |
- Each variable exists only in its own mode.
_itemsin each-item mode (or_itemin all-items mode) raisesNameError. - Dict access only.
it["json"]["name"]orit["json"].get("name").it.json.nameraisesAttributeError: 'dict' object has no attribute 'json'. - No other nodes. There is no
_node/$('Node')equivalent. If you need data from another branch, bring it in with Merge first, or read it in JavaScript. - Webhook payloads are under
["body"]:_items[0]["json"].get("body", {}).get("email"). - Binary data isn't covered here. Read and write binary in JavaScript (see n8n-binary-and-data).
- pairedItem: returning
_items/_itemkeeps it. When you build new dicts and downstream uses$('Node').item, add"pairedItem": {"item": i}to each returned item. That isn't verified on native Python yet, so test-run before relying on it. - Missing keys: prefer
.get(key, default).row["missing"]raisesKeyError.
Migration table for legacy code:
| Legacy (Pyodide) | Native |
|---|---|
_input.all() | _items |
_input.first()["json"] | _items[0]["json"] (guard if _items) |
_input.item / _json | _item / _item["json"] |
_node["X"] | not available: Merge upstream, or use JS |
_now, _today | not available: pass {{ $now.toISO() }} in via Edit Fields, or use JS |
_jmespath(data, q) | not available: $jmespath in an expression, or a comprehension |
item.json.field | item["json"]["field"] |
Imports: blocked by default
Every import (standard library and third-party) is checked against an allowlist before the code runs. The default allowlist is empty, so even import json rejects the whole node:
Security violations detected
Line 1: Import of standard library module 'json' is disallowed. Allowed stdlib modules: none- n8n Cloud: no imports at all.
- Self-hosted: the admin can allowlist modules in the task-runner config (see n8n-self-hosting →
TASK_RUNNERS.md). Some instances therefore allowjson,datetimeandre, and most don't. - Default to import-free code. If an import would really help, confirm it first: a one-line test node
import json+return [{"json": {"ok": True}}], or ask the user. Never assume. - Without imports: parse JSON strings upstream (
{{ JSON.parse($json.payload) }}in Edit Fields). Do date math in expressions (Luxon) or JS. ISO-8601 strings still compare and sort correctly as plain strings. Use the Crypto node for hashing. requests,pandasandnumpyare never available unless the admin built a custom runner image. Use the HTTP Request node for HTTP.
Sandbox limits (these fail even without imports)
| You write | What happens | Use instead |
|---|---|---|
eval, exec, compile, open, input, type, getattr, setattr, hasattr, vars, dir, globals, locals, object, memoryview, breakpoint | NameError: name 'type' is not defined (runtime) | isinstance(x, dict); key in d / d.get(key) |
class Foo: ... | __build_class__ not found (runtime) | dicts + functions |
x.__class__, "{0.__class__}".format(x), __import__("json") | Security violations detected (whole node rejected before running) | — |
global counter inside a function | NameError: name 'counter' is not defined, because your code runs inside a wrapper function | nonlocal counter |
Everything else in plain Python works (verified): comprehensions, generators, lambdas, closures, recursion, try/except, f-strings / .format() / %, sorted/min/max/sum/any/all/ enumerate/zip/round, sets, isinstance, print() (output goes to the browser console).
Return shapes (verified)
Observed on n8n 2.38.5. The auto-wrapping and passthrough behaviours below are undocumented and could change in a later release. Re-check with a test run after upgrading n8n.
Run Once for All Items
| Return | Result |
|---|---|
[{"json": {...}}, ...] | canonical, N items |
[{...}, ...] (plain dicts) | auto-wrapped under json, N items |
{"json": {...}} or a single plain dict | 1 item |
_items (mutated in place) | passthrough with your changes |
None / no return | error Cannot read properties of null (reading 'json') |
Run Once for Each Item
| Return | Result |
|---|---|
{"json": {...}}, a plain dict, or _item | 1 item |
None | the item is dropped (a built-in filter) |
| a list | error A 'json' property isn't a dictionary [item 0] |
Value conversion on output: tuple becomes a list, set becomes the string "{1, 2}", and a datetime becomes str(dt) ("2026-09-16 10:18:00.025792", not ISO). Convert explicitly (sorted(s), dt.isoformat()).
Prefer the explicit [{"json": ...}] in all-items mode and {"json": ...} in each-item mode. The auto-wrapping works, but the explicit shape makes the intent obvious to the next reader.
Errors and onError
A plain raise ValueError("bad row") fails the node with that message. When the node has an onError continue mode, the three failure kinds behave differently (verified):
| Failure | continueErrorOutput | continueRegularOutput |
|---|---|---|
Runtime exception (raise, KeyError, NameError, denied builtin) | {"error": "<message>"} on the error output (main[1]) ✅ | {"error": "<message>"} on the main output |
Static rejection (Security violations detected: import, dunder) | node marked failed, but the input items, unchanged, go out the success output; main[1] stays empty | input items, unchanged, on the main output |
Bad return shape (list in each-item mode, None in all-items mode) | same: unchanged input on the success output | same |
The last two are silent-data traps: downstream nodes receive unprocessed input as if the code had run, and the execution still shows success. No error branch catches them. Prevent them (no imports unless confirmed, correct return shape) and confirm with a real test run, checking the Code node's status and output (see n8n-error-handling).
Other messages:
Python runner unavailable: Python 3 is missing from this system: the self-hosted instance has no Python task runner (the stock image ships none). It's an infrastructure problem, not a code problem. See n8n-self-hosting →TASK_RUNNERS.md.validate_node/validate_workflowcatch a few Python mistakes (import requests, missingreturn,return None). They do not catch_input/_json, dot access, blocked stdlib imports, dunder access or classes. A test execution is the only reliable check.
Performance
Each Python Code node costs roughly 0.4 s (about 1 s when the runner is cold), noticeably more than a JS Code node or an expression. Process lists in Run Once for All Items mode rather than per item, and don't chain several small Python nodes where one would do.
Checklist
- [ ] The user actually asked for Python. Otherwise JS, an expression, or a native node.
- [ ] Mode matches the variable:
_items(all items) /_item(each item). - [ ] Only dict access. No
_input,_json,_node,_nowor_jmespath. - [ ] No
importunless confirmed allowlisted on this instance. - [ ] No classes,
type(),getattr/hasattror dunders.nonlocalinstead ofglobal. - [ ] Return shape fits the mode. Sets and datetimes converted explicitly.
- [ ] Ran a real test execution and inspected the output items (validation alone won't catch the traps above).
Reference
- COMMON_PATTERNS.md: 12 import-free patterns, each verified on a live n8n instance (filter, aggregate, group, dedupe, top N, flatten, validate, drop items, text report, safe nested access, running totals, ISO timestamps).
Related skills
- n8n-code-javascript: the default for Code nodes, with all n8n helpers.
- n8n-expression-syntax:
$jmespath, Luxon and the transform gatekeeper, often a better fit than any Code node. - n8n-code-tool: Python in the AI-agent Custom Code Tool (
_query, returns a string). - n8n-error-handling: wiring error outputs; the passthrough trap above.
- n8n-self-hosting: enabling the Python task runner and allowlisting modules.
Archivos 3 archivos
Instálala a mano
Deska ejecuta esto por ti, para cada agente que elijas. Para ejecutarlo tú en una terminal:
npx skills add czlonkowski/n8n-skills --skill n8n-code-pythonMás de czlonkowski
- n8n-workflow-patternsczlonkowskiProven workflow architectural patterns from real n8n workflows. Use when building new workflows, designing workflow structure, choosing workflow patterns, planning workflow architecture, or asking about webhook processing, HTTP API integration, database operations, AI agent workflows, batch processing, or scheduled tasks. Always consult this skill when the user asks to create, build, or design an n8n workflow, automate a process, or connect services — even if they don't explicitly mention 'patterns'. Covers webhook, API, database, AI, batch processing, and scheduled automation architectures. Also use when optimizing a slow workflow or speeding up large-item-count processing (node count, batchSize, all-items vs per-item).11.2 k instalaciones en skills.sh0
- n8n-mcp-tools-expertczlonkowskiExpert guide for using n8n-mcp MCP tools effectively. Use when searching for nodes, validating configurations, accessing templates, managing workflows, organizing workflows into folders, managing credentials, auditing instance security, or using any n8n-mcp tool. Provides tool selection guidance, parameter formats, and common patterns. IMPORTANT — Always consult this skill before calling any n8n-mcp tool — it prevents common mistakes like wrong nodeType formats, incorrect parameter structures, and inefficient tool usage. If the user mentions n8n, workflows, nodes, or automation and you have n8n MCP tools available, use this skill first.7,900 instalaciones en skills.sh0
- n8n-node-configurationczlonkowskiOperation-aware node configuration guidance. Use when configuring nodes, understanding property dependencies, determining required fields, choosing between get_node detail levels, or learning common configuration patterns by node type. Always use this skill when setting up node parameters — it explains which fields are required for each operation, how displayOptions control field visibility, and when to use patchNodeField for surgical edits vs full node updates.7,200 instalaciones en skills.sh0
- n8n-code-javascriptczlonkowskiWrite JavaScript code in n8n Code nodes. Use when writing JavaScript in n8n, using $input/$json/$node syntax, making HTTP requests with this.helpers / the $helpers global, working with dates using DateTime, troubleshooting Code node errors, choosing between Code node modes, or doing any custom data transformation in n8n. Always use this skill when a workflow needs a Code node — whether for data aggregation, filtering, API calls, format conversion, batch processing logic, or any custom JavaScript. Covers SplitInBatches loop patterns, cross-iteration data, pairedItem, and real-world production patterns. Also use when asked why a Code node or workflow is slow, which execution mode is faster, or how to cut per-item overhead on large datasets. EXCEPTION — for the AI-agent-callable Custom Code Tool (@n8n/n8n-nodes-langchain.toolCode, a tool attached to an AI Agent), use the n8n-code-tool skill instead; it has a different runtime contract.6,800 instalaciones en skills.sh0
- n8n-validation-expertczlonkowskiInterpret validation errors and guide fixing them. Use when encountering validation errors, validation warnings, false positives, operator structure issues, or need help understanding validation results. Also use when asking about validation profiles, error types, the validation loop process, or auto-fix capabilities. Consult this skill whenever a validate_node or validate_workflow call returns errors or warnings — it knows which warnings are false positives and which errors need real fixes.6,400 instalaciones en skills.sh0