Question 1

How do you connect to an API that has no n8n node?

Accepted Answer

Naming the node is not enough — describe the process. "With the HTTP Request node. In practice I do this: 1. Import the cURL example. If the API docs or Postman have a working curl command, I import it into the node — method, URL, headers and body fill in automatically. Faster than typing and free of the usual typos. 2. Move the key into a credential. After the import the token sits in a node parameter; I lift it into a credential via Generic Credential Type, because node parameters end up in workflow exports. 3. Send one request and look at the response — I cannot build the next step without seeing the field names and shape. 4. Set up pagination and limits — the node's Pagination option if the API returns many records, and Batching if there is a rate limit. One nuance: when the service has a node but not the operation I need, I pick Predefined Credential Type in HTTP Request — reusing the existing credential instead of setting authentication up again. n8n itself recommends that."

Question 2

An API returns 10,000 records, 100 per page. How do you fetch them all?

Accepted Answer

Know both approaches and say when you pick each. "My first choice is the HTTP Request node's own Pagination option. It has two modes: incrementing a parameter on each request (page, offset), or following the next URL in the response. The node exposes helper variables — $pageCount for how many pages have been fetched and $response for the response — so you can express the stop condition too. Before building it I always send one request and look at the response shape: does the API give the next page's URL, or does it just expect a page number? Is there a cap on page size? If the scheme is non-standard — a cursor split between a header and the body, say — I build the loop by hand with Loop Over Items. Either way I watch two things: memory (holding 10,000 records in one execution is heavy, so I trim unneeded fields early) and rate limits (an interval between batches)."

Question 3

How do you make a workflow resilient against an unreliable external API?

Accepted Answer

This tests operational thinking. Answer in layers. 1. Timeout. I set the node's Timeout option to a realistic value. Left at the default, a hung request can block the whole execution. 2. Retry. n8n offers per-node retry. But it only makes sense for transient failures: 429, 5xx, network errors. Retrying a 400 or 401 is pointless — the request or the credential is wrong. 3. Different reactions per status code. I turn on "Never Error" and "Include Response Headers and Status" and branch with a Switch: 429 → wait and retry, 404 → mark the record and carry on, 5xx → raise an alert. 4. Tolerating partial failure. If 3 of 100 requests fail I do not want the whole workflow to stop. Using per-node continue-on-error I route the failures down a separate path and report on them afterwards. 5. Visibility. Failures have to be counted. Without an error workflow and a tracked failure rate, you learn the API degraded only when a user complains."

Option	What it does	When you need it
Predefined Credential Type	Uses the ready-made credential type of a service n8n knows	The service has a node but not the operation you need — this is n8n's recommended route
Generic Credential Type	Generic types: header, basic, bearer, query, OAuth2	Your own or an unknown API
Pagination	Walks a paginated response automatically: by incrementing a parameter or following the next URL in the response	When the API returns more than a page of records
Batching	Splits requests into batches with an interval between them	When the API has a rate limit
Timeout	How long to wait for the response to start (milliseconds)	A slow or unreliable service
Never Error	Treats any status code as success	When you want to handle error codes yourself
Include Response Headers and Status	Returns headers and the status code as well as the body	When branching on the status code, or reading rate-limit headers

HTTP Request: connecting to any API

📚 Sources and documentation