How to Use the AI/LLM

LLM

Click to read the full definition in our AI & Automation Glossary.

Node in n8n (Claude/Anthropic)

The Anthropic node in n8n gives you direct access to Claude models inside your workflows. This means you can automate client intake forms, generate case summaries, draft engagement letters, or analyze contracts without leaving your automation stack.

This guide shows you exactly how to configure the node, which parameters matter, and how to build production-ready workflows that won't waste tokens or produce garbage output.

Required:

• Active n8n instance (cloud or self-hosted version 1.0+)
• Anthropic API key from console.anthropic.com
• Basic understanding of n8n workflow canvas

Get your API

1. Sign up at console.anthropic.com
2. Navigate to API

   API

   Click to read the full definition in our AI & Automation Glossary.

   Keys section
3. Click "Create Key"
4. Copy the key immediately (it only displays once)
5. Set usage limits under Settings > Billing to avoid surprise charges

Add the node to your workflow:

1. Open your n8n workflow canvas
2. Click the + button to add a node
3. Search for "Anthropic" or "Claude"
4. Select "Anthropic Chat Model" (not the legacy "Anthropic" node)

Connect your API

1. Click the "Credential to connect with" dropdown
2. Select "Create New Credential"
3. Paste your API

   API

   Click to read the full definition in our AI & Automation Glossary.

   key in the "API

   API

   Click to read the full definition in our AI & Automation Glossary.

   Key" field
4. Name it something memorable like "Anthropic Production Key"
5. Click "Save"

The credential is now available across all workflows in your n8n instance.

The Anthropic node has six parameters that control output quality and cost. Here's what each one does and when to adjust it.

Model Selection:

• claude-3-5-sonnet-20241022: Best balance of speed, cost, and quality. Use this for 90% of tasks.
• claude-3-opus-20240229: Highest quality, slowest, most expensive. Use for complex legal analysis or high-stakes client communications.
• claude-3-haiku-20240307: Fastest and cheapest. Use for simple classification, data extraction, or high-volume tasks.

Prompt (required): This is your instruction to Claude. Be specific. Bad: "Summarize this." Good: "Extract client name, matter type, and key deadlines from this intake form. Return as JSON."

You can reference data from previous nodes using expressions: {{ $json.email_body }}

Max Tokens:

• Controls maximum response length
• 1 token ≈ 4 characters in English
• Default is 1024 (about 750 words)
• Set to 4096 for long-form content
• Set to 256 for short classifications or extractions

Temperature (0.0 to 1.0):

• Controls randomness and creativity
• 0.0 = deterministic, consistent output (use for data extraction, classification)
• 0.7 = balanced creativity (use for drafting emails, summaries)
• 1.0 = maximum variation (use for brainstorming, creative content)

Top P (0.0 to 1.0):

• Alternative to temperature for controlling randomness
• 0.9 is a safe default
• Don't adjust both temperature and top_p simultaneously

Stop Sequences:

• Optional array of strings that halt generation
• Example: ["---END---", "\n\n\n"]
• Useful for structured output or preventing runaway responses

Here's a complete workflow that processes client intake emails and extracts structured data.

Workflow structure:

1. Email Trigger (Gmail, Outlook, or IMAP)
2. Anthropic Chat Model node
3. Set node (to structure the output)
4. Airtable/Google Sheets node (to store results)

Configure the Anthropic node:

Prompt:

Extract the following information from this client intake email:
- Client full name
- Company name (if mentioned)
- Matter type (litigation, M&A, employment, real estate, other)
- Urgency level (high, medium, low)
- Key dates mentioned
- Budget mentioned (if any)

Email content:
`{{ $json.body }}`

Return your response as valid JSON with these exact keys: client_name, company, matter_type, urgency, dates, budget. If information is not present, use null.

Settings:

• Model: claude-3-5-sonnet-20241022
• Max Tokens: 512
• Temperature: 0.2
• Top P: 0.9

Expected output:

{
  "client_name": "Sarah Chen",
  "company": "TechStart Inc",
  "matter_type": "M&A",
  "urgency": "high",
  "dates": ["2024-03-15 board meeting", "2024-03-30 closing deadline"],
  "budget": "$50,000-75,000"
}

Problem: Claude returns markdown formatting instead of clean JSON

Solution: Add this to your prompt:

Return ONLY the JSON object. Do not include markdown code blocks, explanations, or any text outside the JSON structure.

Problem: Inconsistent field names or structure

Solution: Provide an example in your prompt:

Example output format:
{
  "client_name": "John Smith",
  "matter_type": "litigation",
  "urgency": "medium"
}

Problem: Response gets cut off mid-sentence

Solution: Increase max_tokens or add a completion check. Insert a Code node after Anthropic:

if ($json.finish_reason !== 'end_turn') {
  throw new Error('Response truncated - increase max_tokens');
}
return $input.all();

Use Case 1: Contract Clause Extraction

Workflow: PDF → Extract Text → Anthropic → Database

Anthropic Configuration:

• Model: claude-3-5-sonnet-20241022
• Max Tokens: 2048
• Temperature: 0.1

Prompt:

Analyze this contract and extract:
1. Termination clauses (section and exact text)
2. Liability caps (amounts and conditions)
3. Indemnification provisions
4. Governing law and jurisdiction
5. Notice requirements

Contract text:
`{{ $json.contract_text }}`

Format as JSON with arrays for each category. Include section references.

Use Case 2: Client Email Triage and Routing

Workflow: Email Trigger → Anthropic → Switch Node → Route to Teams

Anthropic Configuration:

• Model: claude-3-haiku-20240307 (fast and cheap for classification)
• Max Tokens: 128
• Temperature: 0.0

Prompt:

Classify this email into exactly one category:
- URGENT_LITIGATION (active lawsuit, court deadline, emergency motion)
- URGENT_COMPLIANCE (regulatory deadline, audit request)
- NEW_MATTER (new client, new engagement)
- EXISTING_MATTER (ongoing work, routine update)
- BILLING (invoice question, payment issue)
- ADMINISTRATIVE (scheduling, general inquiry)

Email subject: `{{ $json.subject }}`
Email body: `{{ $json.body }}`

Return only the category name, nothing else.

Use Case 3: Engagement Letter Generator

Workflow: Form Submission → Anthropic → Google Docs → Email

Anthropic Configuration:

• Model: claude-3-5-sonnet-20241022
• Max Tokens: 3072
• Temperature: 0.4

Prompt:

Draft an engagement letter for a law firm with these details:

Client: `{{ $json.client_name }}`
Matter: `{{ $json.matter_description }}`
Scope: `{{ $json.scope_of_work }}`
Fee Structure: `{{ $json.fee_arrangement }}`
Key Team Members: `{{ $json.team_members }}`

Include:
1. Scope of representation (specific and limited)
2. Fee arrangement and billing terms
3. Client responsibilities
4. Conflicts disclosure
5. Termination provisions
6. Standard disclaimers

Use professional but accessible language. Format with clear section headers.

Estimate costs before deploying:

• Claude 3.5 Sonnet: $3 per million input tokens, $15 per million output tokens
• Average client email (500 words) = ~650 tokens input
• Average extraction response = ~200 tokens output
• Cost per email processed: ~$0.005

Reduce token usage:

1. Truncate input text to relevant sections only
2. Use Haiku for simple tasks (10x cheaper than Opus)
3. Set conservative max_tokens limits
4. Cache system prompts when using the same instructions repeatedly

Add usage monitoring: Insert a Code node after Anthropic to log token usage:

const usage = $json.usage;
const cost = (usage.input_tokens * 0.000003) + (usage.output_tokens * 0.000015);

return [{
  json: {
    workflow_id: $workflow.id,
    tokens_used: usage.input_tokens + usage.output_tokens,
    estimated_cost: cost,
    timestamp: new Date().toISOString()
  }
}];

Send this data to a Google Sheet or database for monthly cost tracking.

Add retry logic for API

1. Click the Anthropic node settings (gear icon)
2. Enable "Retry On Fail"
3. Set "Max Tries" to 3
4. Set "Wait Between Tries" to 5000ms

Handle rate limits: Anthropic enforces rate limits based on your tier. If you hit limits, add a Wait node before the Anthropic node:

Wait Time: 1000ms (1 second between requests)

Validate output structure: Add a Code node after Anthropic to verify JSON structure:

const response = $json.response;
let parsed;

try {
  parsed = JSON.parse(response);
} catch (e) {
  throw new Error('Invalid JSON response from Claude');
}

const required = ['client_name', 'matter_type', 'urgency'];
for (const field of required) {
  if (!parsed[field]) {
    throw new Error(`Missing required field: ${field}`);
  }
}

return [{ json: parsed }];

The Anthropic node transforms n8n from a simple automation tool into an intelligent document processor. Start with the Sonnet model for general tasks, use Haiku for high-volume classification, and reserve Opus for complex analysis where accuracy is critical.

Your first workflow should be simple: email in, structured data out. Once that works reliably, expand to contract analysis, document generation, and client communication drafting.

Monitor your token usage religiously. A poorly configured workflow can burn through your API