Agent node

Execute AI-powered tasks using large language models with structured outputs, memory, and MCP server integrations

The Agent node brings AI into your workflows using large language models (LLMs). It processes prompts with dynamic data from previous nodes, supports structured JSON outputs, maintains conversation memory across executions, and can connect to MCP servers for extended capabilities.

When to use

• Analyze and extract insights from enterprise data (invoices, contracts, support tickets)
• Generate summaries, recommendations, or classifications based on workflow data
• Transform unstructured text into structured JSON for downstream processing
• Build conversational workflows that maintain context across multiple interactions
• Extend agent capabilities with MCP servers for external tool access

Configure your LLM API keys under Organization > Settings before using the Agent node.

Input parameters

Core configuration

Model · dropdown · required

Select the AI model to process prompts. Supports models from OpenAI (GPT-4o, GPT-4.1, GPT-5), Google (Gemini 2.0, Gemini 2.5), and Anthropic (Claude 3.5, Claude 3.7, Claude Opus 4).

User Prompt · text area · required

The input text or question to send to the AI model. Supports dynamic variables from previous nodes using {{variable}} syntax.

System Prompt · text area

Instructions that define the agent’s behavior, role, and response style. Sets context before processing the user prompt.

Structured Output · code editor

JSON schema that enforces a specific output format from the agent. When defined, the agent returns responses matching this schema structure.

Structured Output Schema Example:

{
  "name": "analysis_result",
  "strict": true,
  "schema": {
    "type": "object",
    "properties": {
      "summary": {
        "type": "string",
        "description": "Brief summary of the analysis"
      },
      "confidence_score": {
        "type": "number",
        "description": "Confidence level from 0 to 1"
      },
      "categories": {
        "type": "array",
        "items": { "type": "string" },
        "description": "Relevant categories identified"
      }
    }
  }
}

MCP servers

MCP Servers · code editor

Configure Model Context Protocol servers to extend the agent’s capabilities with external tools and data sources.

MCP Server Configuration Example:

[
  {
    "name": "WeatherMCP",
    "url": "https://api.weather.com/v3/weather/daily",
    "headers": {
      "Content-Type": "application/json"
    },
    "query_params": {
      "apiKey": "your_api_key"
    }
  }
]

Memory

Enable memory to maintain conversation context across multiple agent interactions within a workflow or across workflow executions.

Store Chat History · boolean

Enable to save conversation history for multi-turn interactions.

Chat History Identifier · string

Unique identifier to group related conversations. Use dynamic variables to create user-specific or session-specific histories.

Recent Messages to Include · number

Number of previous messages to include in the agent’s context. Default: 10

Output

The node output structure is determined by the Structured Output schema you define. If no schema is specified, the agent returns a text response.

Example Output (with structured schema):

{
  "summary": "The vendor invoice contains 3 line items totaling $15,420.00",
  "confidence_score": 0.95,
  "categories": ["procurement", "accounts_payable", "vendor_management"]
}

Error Response:

{
  "status": "Errored",
  "node_id": "20",
  "node_name": "Agent",
  "body": "No API Key is provided for LLM Node, Please provide API Key in Org Settings"
}

Adding to your workflow

1. Add the Agent node

   In the workflow editor, click Add Node and select Agent from the utility nodes section.

2. Select a model

   Choose an AI model from the dropdown. Different models have varying capabilities, speed, and cost characteristics.

3. Configure prompts

   Enter the User Prompt with your query or instruction. Use {{variable}} syntax to include data from previous nodes. Optionally add a System Prompt to define the agent’s behavior.

4. Define structured output (optional)

   If you need the response in a specific JSON format, define a JSON schema in the Structured Output field.

5. Configure MCP servers (optional)

   Add MCP server configurations to extend the agent’s capabilities with external tools.

6. Enable memory (optional)

   Click Enable under Memory to maintain conversation history. Configure the Chat History Identifier and Recent Messages to Include.

7. Test the node

   Click the Run tab to execute a test and verify the agent’s response.

Examples

Analyze SAP Purchase Order

Extract key information from an SAP purchase order and classify it for routing.

Configuration:

Field	Value
Model	openai/gpt-4o
User Prompt	Analyze this SAP purchase order and extract key details: {{sap_po_response}}

System Prompt:

    You are a procurement analyst. Extract purchase order details and classify by urgency and department. Be concise and accurate.

Structured Output:

    {
      "name": "po_analysis",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "po_number": { "type": "string" },
          "vendor_name": { "type": "string" },
          "total_amount": { "type": "number" },
          "currency": { "type": "string" },
          "urgency": { "type": "string", "enum": ["low", "medium", "high", "critical"] },
          "department": { "type": "string" },
          "requires_approval": { "type": "boolean" }
        }
      }
    }

Output:

    {
      "po_number": "4500012345",
      "vendor_name": "Acme Industrial Supplies",
      "total_amount": 24500.00,
      "currency": "USD",
      "urgency": "high",
      "department": "Manufacturing",
      "requires_approval": true
    }

Classify NetSuite Support Tickets

Automatically categorize and prioritize incoming support tickets.

Configuration:

Field	Value
Model	anthropic/claude-3.7-sonnet
User Prompt	Classify this support ticket: {{netsuite_case.description}}

System Prompt:

    You are a support ticket classifier. Analyze the ticket content and determine category, priority, and suggested routing. Use only the predefined categories.

Structured Output:

    {
      "name": "ticket_classification",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "category": { 
            "type": "string", 
            "enum": ["billing", "technical", "account", "feature_request", "general"] 
          },
          "priority": { 
            "type": "string", 
            "enum": ["low", "medium", "high", "urgent"] 
          },
          "sentiment": { 
            "type": "string", 
            "enum": ["positive", "neutral", "negative", "angry"] 
          },
          "route_to": { "type": "string" },
          "summary": { "type": "string" }
        }
      }
    }

Output:

    {
      "category": "billing",
      "priority": "high",
      "sentiment": "negative",
      "route_to": "billing_team",
      "summary": "Customer disputing invoice charges for Q4 services"
    }

Generate Tipalti Payment Summary

Create a human-readable summary of payment batch status.

Configuration:

Field	Value
Model	google/gemini-2.5-flash
User Prompt	Summarize this payment batch for the finance team: {{tipalti_batch_response}}

System Prompt:

    You are a finance assistant. Create clear, actionable summaries of payment batches. Highlight any issues or items requiring attention.

Structured Output:

    {
      "name": "payment_summary",
      "strict": true,
      "schema": {
        "type": "object",
        "properties": {
          "batch_id": { "type": "string" },
          "total_payments": { "type": "number" },
          "total_amount": { "type": "number" },
          "successful": { "type": "number" },
          "failed": { "type": "number" },
          "pending_review": { "type": "number" },
          "executive_summary": { "type": "string" },
          "action_items": { 
            "type": "array", 
            "items": { "type": "string" } 
          }
        }
      }
    }

Output:

    {
      "batch_id": "BATCH-2024-0142",
      "total_payments": 47,
      "total_amount": 128450.00,
      "successful": 44,
      "failed": 2,
      "pending_review": 1,
      "executive_summary": "Payment batch 94% successful. 2 failures due to invalid bank details, 1 payment pending compliance review.",
      "action_items": [
        "Contact vendor VEND-892 to update bank account information",
        "Contact vendor VEND-1204 to verify routing number",
        "Review payment to VEND-445 for compliance approval"
      ]
    }

Conversational Agent with Memory

Build a multi-turn conversation that maintains context across interactions.

Configuration:

Field	Value
Model	openai/gpt-4.1
User Prompt	{{user_message}}
Store Chat History	Enabled
Chat History Identifier	{{user_id}}_support_session
Recent Messages to Include	10

System Prompt:

    You are a helpful enterprise support assistant. You have access to the user's account information and can help with integration questions. Maintain context from previous messages in the conversation.

Conversation Flow:

Turn 1:

• User: “I’m having trouble with my SAP integration”
• Agent: “I can help with SAP integration issues. Could you describe what specific problem you’re experiencing?”

Turn 2:

• User: “The purchase orders aren’t syncing”
• Agent: “I see you’re having PO sync issues. Let me check a few things. Are you seeing any error messages in the workflow logs?”

Turn 3:

• User: “Yes, it says authentication failed”
• Agent: “Based on our conversation, it sounds like your SAP credentials may have expired. You can update them under Settings > Credentials > SAP. Would you like me to guide you through the process?”

Troubleshooting

API Key Errors

Problem	Solution
No API Key provided	Configure LLM API keys under Organization > Settings.
Invalid API key	Verify the API key is correct and has not expired. Regenerate if needed.
Rate limit exceeded	Check your LLM provider’s rate limits. Consider using a different model or adding delays between requests.

Response Errors

Problem	Solution
Response doesn’t match schema	Review your structured output schema for errors. Ensure property types and constraints are valid JSON Schema.
Empty or incomplete response	Check if the prompt is clear and provides sufficient context. Verify input variables are resolving correctly.
Timeout errors	Reduce prompt complexity or use a faster model variant. Check if the model is experiencing high load.

Memory Issues

Problem	Solution
Conversation history not persisting	Verify Store Chat History is enabled and Chat History Identifier is consistent across executions.
Agent losing context	Increase Recent Messages to Include value. Check if the identifier is unique per conversation.
Unexpected context in responses	Verify the Chat History Identifier is creating separate histories for different users or sessions.

MCP Server Errors

Problem	Solution
MCP server connection failed	Verify the server URL is correct and accessible. Check authentication headers and query parameters.
Tool not available	Ensure the MCP server exposes the required tools. Review server documentation for available capabilities.

Use a Custom Code node before the Agent node to prepare and format complex input data for better prompt quality.

Next steps

Transform node

Process agent outputs with JSONata.

Rule node

Add conditional logic based on agent responses.

Loop node

Process multiple items with the agent.

What is MCP?

Understand how agents interact with enterprise systems.