Generate Python Code for n8n Nodes

Provides expert guidance for writing Python inside n8n's Code node (Beta), anchored on the single most important limitation: no external library imports are available - requests, pandas, numpy, scipy, and BeautifulSoup all raise ModuleNotFoundError; only the Python standard library works (json, datetime, re, base64, hashlib, urllib.parse, math, random, statistics). It documents two execution modes - "Run Once for All Items" (default, code executes once, data via _input.all(), best for aggregation and batch transforms) and "Run Once for Each Item" (_input.item, executes per item, slower on large datasets) - and two Python variants: Python (Beta), which exposes n8n helper syntax (_input, _json, _node, _now, _today, _jmespath()) and is the recommended default, versus Python (Native), which only offers plain _items/_item with no helpers. It also covers four data-access patterns - _input.all() for batches, _input.first() for single objects, _input.item in each-item mode, _node["NodeName"]["json"] to pull data from a specific upstream node - and five worked patterns: data transformation via list comprehension, filtering and aggregation, regex-based string extraction, field validation, and statistical analysis via the statistics module (mean, median, stdev).

return [{
    "json": {
        "field1": value1,
        "field2": value2
    }
}]

Use Python specifically when you need the statistics module, are significantly more comfortable with Python syntax, or your logic maps naturally to list comprehensions - the skill is explicit that JavaScript is the recommended choice for 95% of use cases, because it has full n8n helper access ($helpers.httpRequest()), the Luxon DateTime library for advanced date handling, and no external-library ceiling. HTTP requests, web scraping, and pandas/numpy-style data analysis are all workarounds-only in Python: use an HTTP Request node, or JavaScript's $helpers.httpRequest(), for HTTP; HTTP Request plus an HTML Extract node instead of BeautifulSoup; and the statistics module or manual list/dict math instead of pandas. Simple field mapping, basic filtering, or simple conditionals should skip Code nodes entirely in favor of the Set, Filter, or IF/Switch nodes.

Input is upstream node data reached via _input.all(), _input.first(), _input.item, or _node[...]; webhook payloads are nested under _json["body"], not _json directly - the single most common mistake the guide flags. Output has one hard rule: the code must return a list of dictionaries, each shaped {"json": {...}} - returning a bare dict, a list without the json key, or a plain string all fail downstream node execution. Five documented top mistakes, each with a fix: importing an external library, forgetting the return statement, returning the wrong shape, a KeyError from direct dict indexing instead of .get(), and unwrapped webhook body access.

Complements five sibling n8n skills: n8n Expression Syntax (when to use {{ }} expressions in other nodes versus Python code directly), n8n MCP Tools Expert (search_nodes, get_node_essentials("nodes-base.code"), validate_node_operation() for configuring the Code node itself), n8n Node Configuration (mode and language selection), n8n Workflow Patterns (where Code nodes fit in a transformation step), n8n Validation Expert (validating and auto-fixing Code node config), and n8n Code JavaScript (the recommended alternative and migration path).

n8n workflow builders who specifically need Python's standard library or syntax inside a Code node - not those defaulting to Python out of habit, since the skill itself steers most users toward JavaScript - and who need to avoid n8n-specific traps like external-import failures, webhook body nesting, and the mandatory [{"json": {...}}] return shape.