JSON Path Extractor

The JSON Path Extractor node extracts specific values from a JSON payload using JSONPath expressions, with support for single or multi-output extraction.

What does the JSON Path Extractor node do?

The JSON Path Extractor node takes a JSON string as input and returns one or several values selected with JSONPath expressions. It is the bridge between nodes that emit structured JSON (LLMs, API connectors, SERP tools, scrapers) and downstream nodes that need a precise field, an array slice, or a filtered subset.

The node automatically strips ```json / ``` Markdown fences from its input, so you can pipe an LLM output directly without prior cleaning. It also supports {{variableName}} placeholders inside expressions, resolved against the connected inputs.

Common use cases:

• Pull a single field (e.g. $.title) from a JSON object generated by an LLM.
• Flatten an array of objects to one of its keys (e.g. $.results[*].url).
• Produce several named outputs in one node (multi-output mode) to avoid chaining multiple extractors.

Quick setup

Follow these steps to add and configure the JSON Path Extractor node in your workflow:

Configuration parameters

The node combines one mandatory input port with parameters that depend on the chosen extraction mode.

Required fields

Optional fields

What does the node output?

In single mode, the node exposes a single output port Extracted content carrying the result of the JSONPath query. In multi-output mode, one port is generated per expression, named after the name field of the entry.

Usage examples

Example 1: Extract a single field from an LLM output

You ask an LLM to return a JSON summary and you only need the article title for the next step.

Input (from an LLM, with Markdown fences):

```json
{
  "article_title": "How to optimize your SEO in 2024",
  "keyword": "SEO strategy",
  "word_count": 1250
}

**Configuration:**
- **Multi-Output Mode**: off
- **JSON Path Expression**: `$.article_title`

**Output (`Extracted content`):**

How to optimize your SEO in 2024

### Example 2: Multi-output extraction from a SERP payload

You connect a Serper Dev node and want to feed three downstream nodes with three distinct slices of the response in a single extractor.

**Configuration:**
- **Multi-Output Mode**: on
- **JSON Path Expressions**:
  - `name` = `urls`, `expression` = `$.organic[*].link`
  - `name` = `titles`, `expression` = `$.organic[*].title`
  - `name` = `paa`, `expression` = `$.peopleAlsoAsk[*].question`

**Result:** the node exposes three output ports (`urls`, `titles`, `paa`), each carrying the corresponding extracted array, ready to be plugged into separate downstream branches.

## Common issues

  <Accordion title="Error: invalid JSON / JSONDecodeError on the input">
    **Cause:** the upstream node sent a string that is neither valid JSON nor valid JSON5, even after stripping ` ```json ` fences. Frequent with LLMs that mix prose and JSON.

    **Solution:** insert a **Find and Replace** or **HTML Cleaner** node before the extractor to strip the offending characters, or tighten the LLM prompt so it returns JSON only. JSON5 (single quotes, trailing commas) is tolerated as a fallback.
  </Accordion>
  <Accordion title="Error: Input '<name>' is required for expression but not provided">
    **Cause:** your expression contains a `{{variableName}}` placeholder, but no input named `variableName` is connected to the node.

    **Solution:** wire the missing variable as an input on the node, or rename the placeholder to match an existing connected input.
  </Accordion>
  <Accordion title="The expression returns nothing or an empty array">
    **Cause:** the JSONPath does not match the actual structure of the payload (typo in a key, missing array wildcard `[*]`, wrong index).

    **Solution:** run the upstream node once and inspect its raw output, then adjust the path. When available, use the suggestions surfaced by the settings panel for known source nodes.
  </Accordion>
  <Accordion title="Multi-output mode shows a validation error on save">
    **Cause:** at least one expression has an empty `Name`, two expressions share the same name, or no expression was added at all.

    **Solution:** every expression must have a unique non-empty name and a non-empty expression starting with `$`.
  </Accordion>

## Best practices and pitfalls

<Callout type="tip">
  Prefer **multi-output mode** over chaining several JSON Path Extractors on the same JSON payload: one node parses the JSON once, and downstream branches stay readable.
</Callout>
<Callout type="warning">
  **Validate paths against real data.** A path like `$.results[0].title` silently returns nothing if the upstream payload uses `data` instead of `results` or wraps the array one level deeper. Always check a real run output before relying on it in production.
</Callout>

## How does it fit into a workflow?

The JSON Path Extractor is the canonical "shape adapter" between a JSON-emitting node and a downstream consumer that expects a scalar or a flat list.

```mermaid
graph LR
    LLM[LLM node generates JSON] --> FR[Find and Replace<br/>cleans Markdown]
    FR --> JPE[JSON Path Extractor]
    JPE --> Loop[Loop over items]
    Loop --> Next[Final node]

Related nodes