Resources · n8n integration

n8n Auto-fixing Output Parser nodeConfigure Auto-fixing Output Parser in n8n.

A model that answers in the wrong format stops a workflow dead. The n8n Auto-fixing Output Parser node wraps another output parser, an output parser being the sub-node that turns free text into clean data, and asks a chat model to repair the answer when parsing fails. One parameter, 2 required ports.

Try the Auto-fixing Output Parser node on n8nSee the 1 parameters

Verified Trustpilot reviews · AI, automation & growth agency

Why automate

What does the n8n Auto-fixing Output Parser node actually do?

It sits between a root node and a stricter parser. The root node, an AI Agent or a Basic LLM Chain, asks its chat model for an answer. The wrapped parser checks that answer against the expected shape. When that check fails, the Auto-fixing Output Parser does not throw the error back at the workflow: it sends the failed attempt, the parsing rules and the validation error to a chat model, and asks for a corrected version.

Think of a support ticket triage flow. A chat model reads the ticket and has to return a category and a priority. Nine times out of ten it returns clean data. The tenth time it wraps the answer in a sentence of commentary, the Structured Output Parser rejects it, and the run stops. Plugging this node in front of that parser gives the model one chance to correct itself before the workflow gives up.

Second case, a scraping-to-spreadsheet flow. An AI Agent extracts product fields from a page and writes them to Google Sheets. A missing bracket in the model output breaks the mapping. The retry happens inside the AI branch, so the rows keep landing in the right columns.

Third case, an alerting flow. A model classifies incoming messages and a Slack node posts the result. The classification matters more than the exact wording, so paying for one extra model call beats waking someone up over a formatting slip.

When to skip it: n8n marks this node as deprecated and points to the Structured Output Parser instead. On a new workflow, start with the plain parser and a sharper prompt. Recent chat models follow a schema well on their own, and every fix costs a second call to your model provider, who bills its own API calls under its own terms. Reach for the auto-fixing wrapper on an existing flow that already fails on format, or when the output shape is genuinely hard to hit.

Known limits: the node repairs format, not facts. A parsed answer that fits the schema and says something wrong still fits the schema. There is no attempt counter to tune, no branch for the case where the repair also fails, and the node only exposes 1 option. If you want to compare hosting models and cost logic before committing, the n8n review covers that ground, and n8n training goes through agent wiring step by step.

Connections

What do you plug into it?

This node is a sub-node: it never runs on its own, it hangs off a root node and has its own ports, a port being one of the small labelled sockets under a node where another sub-node connects.

Ports (what plugs in)

  • Chat Modelai_languageModelrequired
  • Output Parserai_outputParserrequired

Output (what it plugs into)

  • Output Parserai_outputParser
01Chat ModelThe model that performs the repair. It receives the failed attempt, the parsing rules and the error message, and returns a corrected answer. This is a second call, separate from the one the root node already made.

The model that performs the repair. It receives the failed attempt, the parsing rules and the error message, and returns a corrected answer. This is a second call, separate from the one the root node already made.

Key parameters

  • Required: leave this port empty and the node blocks the run.
  • OpenAI Chat Model: the usual starting point, and the credential for your provider lives on that sub-node, not here.
  • Anthropic Chat Model or Google Gemini Chat Model: fine alternatives when the rest of the workflow already runs on that provider.
Use cases
a repair pass on a category-and-priority answer, where the fix only has to restore the shape of the data.
02Output ParserThe parser being wrapped. It defines what a valid answer looks like, and its failure is what wakes this node up. Without it there is nothing to auto-fix.

The parser being wrapped. It defines what a valid answer looks like, and its failure is what wakes this node up. Without it there is nothing to auto-fix.

Key parameters

  • Required: the node needs a parser to wrap before it can catch anything.
  • Structured Output Parser: describes the expected object, so rejections are precise and easy to repair.
  • Item List Output Parser: for an answer that should come back as a list of items rather than an object.
Use cases
a product extraction that must always return the same keys, wrapped so a stray line of commentary does not kill the run.
Parameters

Which parameter is there to set?

The Auto-fixing Output Parser node has one parameter. For each one: the node as you configure it in n8n, what the parameter changes, and our field notes.

01

Options

options

What you see in n8n

Notes & use cases

The node ships with almost nothing to configure. Everything sits behind a single collection, opened with Add Option, and nothing in it is required: leave it closed and the node uses its built-in repair prompt.

Key parameters

  • Retry Prompt: the prompt template used for fixing the output. It has to include the {instructions}, {completion} and {error} placeholders, which stand for the parsing rules, the failed attempt and the validation error message.
Use cases
add a line to the retry prompt telling the model to answer with the object alone and no commentary, when that is the mistake it keeps making.
Need help

Need help automating Auto-fixing Output Parser with n8n?

A person reads every message.

FAQ

Questions people ask next

01Is the n8n Auto-fixing Output Parser node free to use?
Yes, on the n8n side. The node belongs to n8n's AI package and ships with n8n, so there is nothing to install and nothing extra to pay, on n8n Cloud as well as on a self-hosted instance under the Community Edition and its Sustainable Use licence. A workflow behaves the same either way. What does have a cost is the model itself: every repair is a second call to the chat model you plug into the Chat Model port, and that provider bills its own API calls under its own terms. That is the real trade-off to weigh, not the n8n side.
02What credentials does it need?
None. This node has no credential and no Authentication selector, so there is nothing to connect, nothing to rotate and no account to create for it. The key lives one step away, on the chat model sub-node you attach to the Chat Model port: that sub-node carries the provider credential, and it is the only place an API key appears in this setup. Same story for the wrapped parser, which has no account of its own either. If the run fails on authentication, open the model sub-node, not this one.
03What are the limits of this node?
Two worth knowing. First, n8n flags it as deprecated and recommends the Structured Output Parser instead, so a workflow built today should start there and only add this wrapper if format failures actually show up. Second, the surface is thin: version 1 of the node, 1 option, and that option is only the Retry Prompt. There is no retry count to tune and no separate path for a repair that fails in turn. It also fixes shape, never substance: an answer that parses cleanly can still be wrong.
04What do you need to plug in at minimum?
Both input ports, because both are required. On the Chat Model port, a chat model sub-node such as the OpenAI Chat Model, which is the easiest place to start since its credential is the one you probably already have in n8n. On the Output Parser port, the parser you want to protect, most often the Structured Output Parser, or the Item List Output Parser when the answer should be a list. Then connect the whole thing to a root node: an AI Agent, an AI Agent Tool or a Basic LLM Chain.
05n8n or Make for this kind of AI workflow?
It depends on where you want the workflow to live. Make is hosted by Make, with no self-hosting option, and it bills per operation, so a retry that costs a second model call is also a second billed step. n8n runs on n8n Cloud or on your own server, which matters when the data going through the parser is sensitive, and its cost model does not track the number of steps the same way. Make's visual logic is more linear; the n8n AI nodes work as root nodes with sub-nodes hanging off ports, which is exactly the shape this parser assumes.
Hack'celeration Lab

Get our weekly integration tips.

No spam. Unsubscribe anytime.

Same platform

Other docs you might need on this stack.

Browse all integrations