Migrating from Function Node to Toolkit/MCP/Agent Nodes

​

Overview

• Function Node is now deprecated and displays a warning icon with “Deprecated!” text in the node header
• Creating new Function nodes is no longer possible through the Add Node menu
• Existing Function nodes continue to function but should be migrated
• Toolkit Node, MCP Node, and Agent Node are the recommended replacements
• Hovering over the warning icon shows a tooltip with migration guidance

​

How to Identify Deprecated Function Nodes

• Warning Icon: A yellow/orange warning triangle icon appears in the node header
• “Deprecated!” Text: Text next to the icon clearly indicates the deprecation status
• Tooltip: Hovering over the warning area shows: “This node is deprecated and will be removed in a future version. View Migration Guide” with a link to this guide

• Review pipelines containing Function nodes
• Choose the appropriate replacement: Toolkit, MCP, or Agent node
• Follow migration steps to update your workflow
• Test the migrated pipeline thoroughly

• Toolkit Node: Provides enhanced functionality for executing toolkit tools with structured output support and better configuration options
• MCP Node: Dedicated node for Model Context Protocol servers with specialized MCP tool handling
• Agent Node: Enables execution of AI agents with task-based instructions and enhanced capabilities

​

Migration Options

​

Toolkit Node

• Your Function node uses tools from ELITEA toolkits
• You need structured output support
• You want enhanced configuration options for toolkit execution
• You’re working with external service integrations

• Full functionality coverage of deprecated Function Node for toolkits
• Structured output configuration support
• Enhanced input/output mapping capabilities
• Better toolkit tool management

​

MCP Node

• Your Function node uses MCP server tools
• You’re working with Model Context Protocol integrations
• You need MCP-specific tool filtering and configuration
• You want dedicated MCP server support

• Dedicated MCP server support
• MCP-specific tool filtering
• Enhanced MCP tool discovery
• Structured output for MCP operations

​

Agent Node

• Your Function node’s behavior can be delegated to an AI agent
• You need intelligent task execution with natural language instructions
• You want to leverage pre-configured agents with specialized capabilities
• Your workflow requires LLM-based decision making

• Natural language task instructions
• Access to pre-configured agents
• Enhanced AI capabilities
• Flexible input mapping with task and chat_history support

​

Migration Steps: Function → Toolkit

1. Navigate to Pipelines in the main menu
2. Open the pipeline containing the Function node
3. Click the Configuration tab
4. Select the Flow tab to view the pipeline flow diagram
5. Locate the Function node - it will display with a warning icon and “Deprecated!” text in the node header
6. Click on the Function node to open its configuration panel
7. Review its current configuration:
   • Toolkit: Note which toolkit is selected (e.g., “confluence_toolkit”)
   • Tool: Note which specific tool is selected (e.g., “read_page_by_id”)
   • Input: Review input state variables
   • Output: Review output state variables
   • Input Mapping: Document all parameter mappings
   • Interrupt Settings: Note any interrupt configurations

1. In the same Flow tab, click the Add Node button (+ icon)
2. Select Toolkit from the node menu under Execution Nodes
3. The new Toolkit node will appear on the canvas
4. Position it near the deprecated Function node for easy comparison

1. Click on the Toolkit node to open its configuration panel
2. Set the Toolkit field:
   • Select the same toolkit that was selected in the Function node
   • Example: Select “confluence_toolkit” from the dropdown
3. Set the Tool field:
   • After selecting the toolkit, the Tool dropdown will appear
   • Select the same tool that was used in the Function node
   • Example: Select “read_page_by_id”
4. Configure Input:
   • Select the same input state variables that were used in the Function node
   • Example: If Function node had page_id, input, select those same variables
5. Configure Output:
   • Select the same output state variables that were used in the Function node
   • Example: If Function node output to page_content, messages, configure the same
6. Set up Input Mapping:
   • Recreate all parameter mappings from the Function node
   • The Input Mapping section displays parameters required by the selected tool
   • For each parameter, configure the same type and value as in the Function node:
     • Variable: Direct state reference (e.g., page_id)
     • F-String: Formatted string with variables (e.g., Page ID: {confluence_page_id})
     • Fixed: Static value (e.g., 123456789)
7. Configure Structured Output (if needed):
   • Toolkit nodes support structured output configuration (Function nodes did not)
   • Enable if you want to enforce a specific output schema
8. Configure Interrupt Settings:
   • Copy interrupt before/after settings from the Function node
   • Example: Enable “Interrupt after” if it was enabled in the Function node

1. Disconnect the incoming connection to the Function node:
   • Click on the edge connecting to the Function node
   • Press Delete or click the remove icon
2. Connect the Toolkit node:
   • Drag a connection from the previous node to the Toolkit node’s input handle
   • Example: Connect from “Data Input Node” to the new Toolkit node
3. Connect the Toolkit outputs:
   • Drag a connection from the Toolkit node’s output handle to the next node in the workflow
   • Example: Connect to “Results Processor Node”
4. Verify all connections match the original Function node’s flow

state:
  input:
    type: str
  messages:
    type: list
entry_point: Function 1
interrupt_after: []
nodes:
  - id: Function 1
    type: function
    input:
      - input
    input_mapping:
      page_id:
        type: fixed
        value: '3703049'
      skip_images:
        type: fixed
        value: false
    output:
      - messages
    structured_output: false
    tool: read_page_by_id
    toolkit_name: Confluence
    transition: END

state:
  input:
    type: str
  messages:
    type: list
entry_point: Toolkit 1
interrupt_after: []
nodes:
  - id: Toolkit 1
    type: toolkit
    input:
      - input
    input_mapping:
      page_id:
        type: fixed
        value: '3703049'
      skip_images:
        type: fixed
        value: false
    output:
      - messages
    structured_output: false
    tool: read_page_by_id
    toolkit_name: Confluence
    transition: END

1. Click on the Function node to select it
2. Click the Delete icon (trash can) or press Delete key
3. Confirm the deletion
4. The Function node is now removed from the pipeline

1. Click Save to save the pipeline configuration
2. Navigate to the Execution tab
3. Run the pipeline with test inputs
4. Verify that:
   • The Toolkit node executes the same tool as before
   • Input/output handling works correctly
   • All parameter mappings function as expected
   • All workflow connections function properly

​

Migration Steps: Function → MCP

1. Navigate to Pipelines in the main menu
2. Open the pipeline containing the Function node
3. Click the Configuration tab
4. Select the Flow tab to view the pipeline flow diagram
5. Locate the Function node using an MCP server tool - it will display with a warning icon and “Deprecated!” text in the node header
6. Click on the Function node to open its configuration panel
7. Review its current configuration:
   • Toolkit: Note which MCP server is selected (e.g., “GitHub MCP”)
   • Tool: Note which specific MCP tool is selected (e.g., “get_file_contents”)
   • Input: Review input state variables
   • Output: Review output state variables
   • Input Mapping: Document all parameter mappings
   • Interrupt Settings: Note any interrupt configurations

1. In the same Flow tab, click the Add Node button (+ icon)
2. Select MCP from the node menu under Execution Nodes
3. The new MCP node will appear on the canvas
4. Position it near the deprecated Function node for easy comparison

1. Click on the MCP node to open its configuration panel
2. Set the Toolkit field:
   • Select the same MCP server that was selected in the Function node
   • MCP servers appear in the toolkit dropdown with “(MCP)” suffix
   • Example: Select “GitHub MCP” from the dropdown
3. Set the Tool field:
   • After selecting the MCP server, the Tool dropdown will appear
   • Select the same MCP tool that was used in the Function node
   • Example: Select “get_file_contents” from available MCP tools
4. Configure Input:
   • Select the same input state variables that were used in the Function node
   • Example: If Function node had owner, repo, path, select those same variables
5. Configure Output:
   • Select the same output state variables that were used in the Function node
   • Example: If Function node output to file_content, messages, configure the same
6. Set up Input Mapping:
   • Recreate all MCP tool parameter mappings from the Function node
   • The Input Mapping section displays MCP tool parameters
   • For each parameter, configure the same type and value as in the Function node:
     • Variable: Direct state reference (e.g., owner, repo, path)
     • F-String: Formatted string with variables (e.g., {repo_owner}/{repo_name})
     • Fixed: Static value (e.g., ProjectAlita for owner, projectalita.github.io for repo)
7. Configure Structured Output (if needed):
   • MCP nodes support structured output configuration (Function nodes did not)
   • Enable if you want to enforce a specific output schema for MCP responses
8. Configure Interrupt Settings:
   • Copy interrupt before/after settings from the Function node
   • Example: Enable “Interrupt after” if it was enabled in the Function node

1. Disconnect the incoming connection to the Function node:
   • Click on the edge connecting to the Function node
   • Press Delete
2. Connect the MCP node:
   • Drag a connection from the previous node to the MCP node’s input handle
   • Example: Connect from “Repository Input Node” to the new MCP node
3. Connect the MCP outputs:
   • Drag a connection from the MCP node’s output handle to the next node in the workflow
   • Example: Connect to “Content Processor Node”
4. Verify all connections match the original Function node’s flow

state:
  file_content:
    type: str
    value: ''
  file_path:
    type: str
    value: ''
  input:
    type: str
  messages:
    type: list
entry_point: Function 1
interrupt_after: []
nodes:
  - id: Function 1
    type: function
    input:
      - input
      - file_path
    input_mapping:
      branch:
        type: fixed
        value: main
      owner:
        type: fixed
        value: ProjectAlita
      path:
        type: variable
        value: file_path
      repo:
        type: fixed
        value: projectalita.github.io
    output:
      - messages
      - file_content
    structured_output: false
    tool: get_file_contents
    toolkit_name: MCP-GitHub
    transition: END

state:
  file_content:
    type: str
    value: ''
  file_path:
    type: str
    value: tests.md
  input:
    type: str
  messages:
    type: list
entry_point: Mcp 1
interrupt_after: []
nodes:
  - id: Mcp 1
    type: mcp
    input:
      - input
    input_mapping:
      branch:
        type: fixed
        value: main
      owner:
        type: fixed
        value: ProjectAlita
      path:
        type: variable
        value: file_path
      repo:
        type: fixed
        value: projectalita.github.io
    output:
      - file_content
      - messages
    structured_output: false
    tool: get_file_contents
    toolkit_name: MCP-GitHub
    transition: END

1. Click on the Function node to select it
2. Click the Delete icon (trash can) or press Delete key
3. Confirm the deletion
4. The Function node is now removed from the pipeline

1. Click Save to save the pipeline configuration
2. Navigate to the Execution tab
3. Run the pipeline with test inputs
4. Verify that:
   • The MCP node executes the same MCP tool as before
   • Input/output handling works correctly
   • All MCP parameter mappings function as expected
   • MCP server connection is maintained
   • All workflow connections function properly

​

Migration Steps: Function → Agent

1. Navigate to Pipelines in the main menu
2. Open the pipeline containing the Function node
3. Click the Configuration tab
4. Select the Flow tab to view the pipeline flow diagram
5. Analyze the Function node to determine if it’s suitable for agent migration:
   • Good candidates: Functions that perform complex reasoning, content generation, analysis, or decision-making
   • Poor candidates: Simple deterministic tool calls, API operations, or direct data transformations
   • Example good candidate: A Function node that processes user stories and generates summaries
   • Example poor candidate: A Function node that simply calls confluence_toolkit.read_page_by_id with static parameters

1. In the same Flow tab, click the Add Node button (+ icon)
2. Select Agent from the node menu under Interaction Nodes
3. The new Agent node will appear on the canvas
4. Position it near the deprecated Function node for easy comparison

1. Click on the Agent node to open its configuration panel
2. Set the Agent field:
   • Select an appropriate agent from the dropdown that can perform the Function node’s task
   • Agents must be added to the pipeline first (Configuration > Toolkits > Add Agent)
   • Example: Select “Content Analysis Agent” to replace a content processing Function node
3. Configure Input:
   • Select the input states that the agent needs
   • Example: If the Function node used user_data, context, select those state variables
4. Configure Output:
   • Select the output states where agent results should be stored
   • Example: If the Function node output to processed_result, configure the same
5. Set up Input Mapping:
   • Configure the agent task and parameters to replicate Function behavior
   • Common Agent Parameters:
     • task: The main instruction/prompt for the agent
     • chat_history: Conversation history (messages array)
     • Custom Variables: Agent-specific parameters
   • Example Task Mapping:
     task: |
       Analyze the following user data and provide insights:
       
       User Data: {input.user_data}
       Context: {input.context_info}
       
       Perform the same analysis that was previously done by the toolkit function.
       Provide:
       1. Key findings from the data
       2. Recommendations based on the analysis
       3. Summary in the same format as before
     
   • Example Chat History Mapping:
     chat_history: messages
     
6. Configure Interrupt Settings:
   • Copy interrupt before/after settings from the Function node
   • Example: Enable “Interrupt after” if it was enabled in the Function node

1. Disconnect the incoming connection to the Function node:
   • Click on the edge connecting to the Function node
   • Press Delete
2. Connect the Agent node:
   • Drag a connection from the previous node to the Agent node’s input handle
   • Example: Connect from “Data Input Node” to the new Agent node
3. Connect the Agent outputs:
   • Drag a connection from the Agent node’s output handle to the next node in the workflow
   • Example: Connect to “Results Processor Node”
4. Verify all connections match the original Function node’s flow

Data Input Node → Function Node → Results Processor Node

Data Input Node → Agent Node → Results Processor Node

1. Click on the Function node to select it
2. Click the Delete icon (trash can) or press Delete key
3. Confirm the deletion
4. The Function node is now removed from the pipeline

1. Click Save to save the pipeline configuration
2. Navigate to the Execution tab
3. Run the pipeline with test inputs
4. Verify that:
   • The Agent node produces equivalent results to the Function node
   • Input/output handling works correctly
   • Task instructions properly guide the agent
   • All workflow connections function properly
5. Compare outputs: Ensure agent results match expected Function node behavior
6. Adjust task instructions: If results differ, refine the agent task prompt

​

Best Practices

# Toolkit Node - Good Practice
✔️ Good: Complete parameter mapping
  toolkit: confluence_toolkit
  tool: read_page_by_id
  input_mapping:
    page_id:
      type: variable
      value: confluence_page_id
    expand:
      type: fstring
      value: "body.storage,version"
    status:
      type: fixed
      value: current

# Toolkit Node - Poor Practice
✘ Avoid: Missing required parameters
  toolkit: confluence_toolkit
  tool: read_page_by_id
  input_mapping:
    # Missing page_id - will cause errors
    expand:
      type: fixed
      value: body.storage

​

Troubleshooting

Error: Tool 'read_page_by_id' not found in toolkit 'confluence_toolkit'
Solution: Ensure the toolkit name and tool name are correct. Check that the toolkit is properly configured and the tool exists.

Error: Missing required parameter 'page_id' for tool execution
Solution: Add input mapping: 
  page_id:
    type: variable
    value: confluence_page_id

Error: Agent produces incorrect output format
Solution: Refine task instructions to specify exact output format:
  task: |
    Analyze the data and provide results in this exact format:
    1. Key Findings: [list]
    2. Recommendations: [list]
    3. Summary: [paragraph]

​

Comparison Table

​

Related Documentation