92 Pipedream workflows in n8n: what imports and what runs

Pipedream Workflows stops on March 31, 2027. We wanted to know what really happens when real Pipedream workflows land in n8n. So we took every public Pipedream workflow we could find on GitHub, converted each one, imported it into n8n 2.44.5 and ran it.

This page has the results. The raw data, one row per workflow, is on GitHub: aidnurs/pipedream-n8n-migration-data. Run date: October 22, 2026.

The sample

We searched GitHub for files in Pipedream's GitHub Sync format (schema: workflow/2022.04 in workflow.yaml). The search returned 95 files. 3 were test files of our own converter, so we left them out. That leaves 92 workflows from 14 public repositories.

Measure	Value
Workflows	92 (8 of them have no steps)
Steps	391, median 3.5 per workflow, max 17
Code steps	266, or 68% of all steps
Workflows with at least one code step	82 of 92
Python steps	42, in 21 workflows
Workflows that import npm packages	29 (snowflake-sdk in 15)
If/else or switch blocks	0

Most workflows come from three repositories:

Repository	Workflows	Steps	Code steps
mjmena/Trend-Tree	30	172	112
cfjedimaster/General-Pipedream-Stuff	23	84	62
CPL-Compliance/priority-xero	14	81	58
11 other repositories	25	54	34

What the code uses

We searched the code of every workflow for the patterns that matter in n8n. Count of workflows, out of 92:

Pattern in the code	Workflows	In n8n
fetch(	31	Not defined in the Code node
axios	29	npm package; not allowed by default
$.respond(	27	Respond to Webhook node
$.flow.*	20	Early return, Wait node or sub-workflow
AbortController	18	Not defined in the Code node
process.env	15	Not defined; move secrets to credentials
Data stores	12	Data tables
new URL( / URLSearchParams	9	Not defined in the Code node
Node built-ins (crypto, fs…)	9	Blocked until you allow them
Files in /tmp	6	Binary data or a mounted volume

The trigger is stored only as an id in GitHub Sync files. Its prefix still shows the kind: HTTP 50 workflows, schedule 24, deployed source 21, app subscription 8. The URL, the interval and the app options are not in the file. You set them again in n8n.

Import: 84 of 84

We converted the 84 workflows that have steps with our converter and imported each file with n8n import:workflow into the n8nio/n8n:2.44.5 Docker image. All 84 imported with no errors.

Import is the easy part. The report put 64 of them in yellow (works after a few manual items) and 20 in red (at least one step needs hand work). None of the 84 came out green, because GitHub Sync files never carry trigger settings. Our estimate: about 199 hours of work left after conversion, against about 318 hours to rebuild them by hand. That is a median of 2 hours instead of 3.25 per workflow.

Run: the first error of each workflow

Then we ran each workflow once with n8n execute. The container had no network. The CLI cannot start a workflow from a webhook, so we swapped every Webhook and Schedule trigger for a manual trigger. Every run started with no input, no credentials and no network. That is harsh on purpose: it shows the first thing that breaks.

Result	Workflows	What it means
Ran to the end	8
Needs real trigger data	20	The code expects a request body, such as missing 'query' in request body
Python runner missing	16	The plain n8n image has no Python. Self-hosted n8n needs the n8nio/runners sidecar
Module disallowed	13	snowflake-sdk ×4, crypto ×4, date-fns, fast-xml-parser, @atproto/api and others
No network	10	Expected: the network was off
No credentials	8	Expected: you create credentials in n8n
Step left for manual work	6	App actions the converter does not map. Flagged red in the report before import
HTTP Request without a URL	1	The URL was already empty in the Pipedream workflow
Dynamic import()	1	Not supported in the Code node
Timeout after 60 seconds	1

So 38 of the 76 failures come from the test setup: no input, no network, no credentials. Those go away when the workflow runs with real data. The other failures are what you plan your migration around:

• Python. 16 workflows stopped because the default image has no Python. On self-hosted n8n you add the external task runner. On n8n Cloud, Python runs. See Pipedream Python steps in n8n.
• Packages and built-ins. 13 workflows stopped on a blocked module. On self-hosted n8n you install the package in the runner image and allow it. On n8n Cloud only crypto and moment load. See npm packages in n8n.
• No fetch is not defined errors. In our first run on October 21, two workflows failed on it. Our converter now replaces fetch with a small wrapper around n8n's own HTTP helper. On October 22 no workflow failed on fetch.

Which globals the Code node has

We ran a probe workflow in the same n8n 2.44.5 image with the default JavaScript task runner. You can import it yourself: code-node-globals.json.

Not defined	Available
fetch, URL, URLSearchParams, AbortController, structuredClone, Blob, process, global crypto	require, Buffer, setTimeout, TextEncoder, atob, btoa, FormData, this.helpers.httpRequest

In the plain image, require('crypto'), require('fs'), require('moment') and require('lodash') all failed with "Module … is disallowed". You allow built-ins with NODE_FUNCTION_ALLOW_BUILTIN and npm packages with NODE_FUNCTION_ALLOW_EXTERNAL. To check a single code step, paste it into the free n8n Code node compatibility check.

Limits of this test

• Not a random sample. Public GitHub Sync repos are what people chose to share. Three repos hold 67 of the 92 workflows.
• No branches. None of the 92 uses if/else or switch blocks, so this test says nothing about them.
• Sync files are not exports. A project export (Project Settings → Export Workflows) has a different format. Neither has secrets or connected accounts.
• First error only. A workflow that stopped on missing input may hit another error later.

The data

The GitHub repo aidnurs/pipedream-n8n-migration-data has:

• data/workflows.csv and data/workflows.json: one row per workflow with step counts, patterns found, report result, run result and first error.
• data/sources.csv: the 14 repositories and the commit we read.
• probe/code-node-globals.json: the probe workflow.

The data is free to use under CC BY 4.0. The repo links to each workflow and does not copy its code. We built this test while working on moveworkflows, a paid converter. The report and 1 workflow are free. Full export: $3 per workflow, $1 each after 50, $19 minimum, one-off.