How can I produce clean, unwrapped Markdown and JSON outputs from LLMs?

Below is a summary of some approaches to resolve unwanted wrappers from LLM outputs within n8n:

---

1. Use a Structured Output Parser

For JSON outputs, adding the Structured Output Parser as an attachment point to your AI node can help. Once you enable the Require Specific Output Format option in your AI node, n8n provides an attachment point where you can add a Structured Output Parser node. This node is designed to format the final output from AI agents into a precise structure so that extraneous text (like “Here is the JSON:”) is removed.

Note: The Structured Output Parser is primarily intended for the final output. It’s best used when you can define a clear schema (or example output) so that the output parser knows exactly what to extract.

---

2. Clean Up the Output Using a Function Node

If the LLM output still includes unwanted Markdown formatting or extra text, you can process and clean up the result using a Function node. For example, you can use regular expressions to remove code block markers (e.g., triple backticks) or any unwanted text fragments.

Below is a sample Function node code snippet that:

• Removes Markdown code blocks (e.g., lang … ).
• Removes any preamble text like “Here is the JSON:” if needed.

JAVASCRIPTCopied
// Get the raw output from the previous node
let output = $json.LLM_output;

// Remove Markdown code block wrappers (handles optional language identifiers)
output = output.replace(/^```(?:\w+)?\s*[\r\n]+|[\r\n]+\s*```$/g, '');

// Optionally, remove a starting phrase if present (e.g., "Here is the JSON:")
output = output.replace(/^Here is the JSON:\s*/i, '');

// Trim extra whitespace
output = output.trim();

return { json: { cleanOutput: output } };

How to integrate:

1. After your AI node: pass the raw output into the Function node.
2. Configure the Function node: copy the code above into the Function node, adjusting the regular expressions if your output patterns differ.
3. Continue your workflow: use the cleaned result (cleanOutput) in subsequent nodes.

---

3. Additional Data Transformation Nodes

While not specifically designed for cleaning text wrappers, other data transformation nodes (like Split In, Set, or even Text Replace if available) can assist if your workflow requires additional manipulation of the output.

---

Recap

• Structured Output Parser: Attach it to your AI node (using the “Require Specific Output Format” option) to standardize the final JSON.
• Function Node Cleanup: Use a Function node with JavaScript regex to remove unwanted Markdown/JSON wrappers or extra text.
• Data Transformation Nodes: Use these for further processing if necessary.

By employing one or a combination of these techniques, you can automate the cleanup process and help ensure that your workflows receive consistent, clean outputs.