LlamaParse Platform Node for n8n: Parse, Classify, Extract & Retrieve Documents with AI

What the node does

What changed in v5

What changed in v6

Install the node on your n8n instance

Example workflows

1. Retrievers as agent tools

1. Classify → Extract → Verify

1. Evaluate Parsed Outputs

Get started

We've shipped v5 and v6 of the LlamaParse Platform community node for n8n, and the package is now an officially verified n8n community node. This post covers what the node exposes, what changed across the two major versions, and three concrete workflows you can build with it.

What the node does

The package ships a single node, LlamaParse Platform, which surfaces five LlamaCloud resources behind one resource/operation dropdown. All five share the same LlamaParse API credential.

Resource Input Output

Parse binary file { text } — Markdown or plain text

Classify binary file + { category, description } rules { category, reasons, confidence }

Split binary file + categories one item per segment: { category, confidence, pages }

Extract binary file + Extraction Config ID { result } — serialized JSON matching your schema

Retrieve indexId, query, topK { context: string[] }

Parse runs across four tiers (fast , cost_effective , agentic , agentic_plus ) with optional pinned parser versions. Retrieve targets both the v1 Pipelines API and the v2 Index Retrieval API. Every resource is marked usableAsTool: true , so it can be attached to an n8n AI Agent as a callable tool rather than only living in a linear flow.

What changed in v5

v5 was a foundation rewrite:

Dropped the SDK in favor of direct HTTP calls against the LlamaCloud API, meaning no need for bundling and maintaining the SDK, as well as a smaller, more predictable surface.

Extract migrated from V1 to V2, aligning with the current Extract service product config model.

Configurable API base URL, so region-specific and self-hosted deployments are a credential setting rather than a fork.

Reworked binary file handling so files flow cleanly from any upstream node.

What changed in v6

v6 consolidated the package. Earlier releases shipped several distinct nodes (LlamaParse, LlamaClassify, LlamaExtract, and so on); v6 collapses them into the single LlamaParse Platform node with the resource/operation pattern used by first-party n8n nodes. It also adds index actions for managing and retrieving from indexes, and incorporates fixes from the n8n verification review (linting, packaging, and class-naming corrections).

Install the node on your n8n instance

In order to install the LlamaParse Platform node, head over to Settings → Community nodes on your n8n dashboard, then provide @llamaindex/n8n-nodes-llamacloud as NPM package name and click on the install button, as in the picture:

Example workflows

To use the LlamaParse integration, add a LlamaParse API credential (paste your key from cloud.llamaindex.ai; base URL defaults to https://api.cloud.llamaindex.ai ) and drop the node onto the canvas.

1. Retrievers as agent tools

Because the node is usableAsTool: true , retrieval resources can be wired directly into an AI Agent's tool list:

The agent decides whether to call the retriever tool based on the user’s query. For instance, if a user asks a domain-specific question that requires context not currently available to the agent, the agent can retrieve the relevant information from the Index. The agent invokes the tool, receives retrieved chunks back and reasons over them in the same turn: just like that, you have plugged in an agent into your own knowledge base, and you can easily expand it with more tools and connections.

1. Classify → Extract → Verify

A document-processing pipeline that turns raw files into structured, verified output:

Classify routes each document to the right branch (in the example,invoice vs. contract ), Extract pulls structured fields using a schema tuned for that type, and an LLM node verifies the correctness of the extracted data based on the parsed content of the file. In addition to what is shown in the example, the confidence field from Classify could be a convenient gate for sending low-confidence documents to a manual-review branch.

1. Evaluate Parsed Outputs

Evaluate parsed outputs with different parsing modes:

Chain Parse in Agentic and a Cost Effective mode, then pass the input file through both parsers and collect the results into an LLM prompt that looks like this:

{agentic output}

{cost effectie output}

Determine which tier has produced the highest-quality result

Finally, represent the LLM-as-a-judge verdict as a structured output you can use for evaluation: this is particularly relevant if you are trying to find the best balance between accuracy and latency for your documents.

Get started

You can get started now with the LlamaParse Platform node, by using it directly in the n8n platform or by checking out:

the GitHub repository: github.com/run-llama/n8n-llamacloud

the npm package: @llamaindex/n8n-nodes-llamacloud

the usage guide: website

the LlamaParse Platform: cloud.llamaindex.ai

Related articles