Agent Node List

This chapter covers the full node catalog you can use when building an agentflow on the canvas, along with the input/output and parameter detailed specs for each node. Every node (agentflow building block) registered in the platform is listed by category, so reading it alongside the Creating an Agent node-adding steps helps you quickly decide which node to pick and how to wire it.

For the actual procedure of adding nodes to the canvas, see Creating an Agent · Adding Nodes. This chapter is the reference for the nodes you choose there.

Node Categories

The root category XGen contains 10 function groups, each composed of the nodes listed below. Drag any node onto the canvas to add it to an agentflow.

Environment-specific agents may appear under the Agent group, so the exact list can vary slightly by environment. The catalog below reflects the standard solution environment.

MCP (mcp)

Node	ID	Description
MCP Tool Loader	mcp/MCPLoader	Connect to any MCP (Model Context Protocol) server and load all its tools at once. Pick an active MCP session
Tavily Search MCP	mcp/tavily_search_mcp	Search the web with Tavily — AI-optimized search returning structured results. Domain filter, summary, raw content
Brave Search MCP	mcp/brave_search_mcp	Real-time web search via Brave Search API. Filter by country and period (day/week/month/year), adjust result count
EPG DAUM MCP	mcp/epg_daum_mcp	Fetch Korean home-shopping TV schedules from DAUM. Broadcast times, program names, channel info (cache-supported)
EPG NAVER MCP	mcp/epg_naver_mcp	Fetch Korean home-shopping TV schedules from NAVER. Same features as DAUM EPG, NAVER data source
GitHub MCP	mcp/github_mcp	Manage GitHub repos via natural language — repos, issues, PRs, code search. App auth
GitLab MCP	mcp/gitlab_mcp	Manage GitLab projects via natural language — projects, branches, files, issues, merge requests
Meta Search MCP	mcp/meta_search_mcp	AI auto-finds and crawls relevant sites for comprehensive info. No API key needed
Naver Datalab MCP	mcp/naver_datalab_mcp	Search trend analytics and shopping insights. Popular terms, search-volume trends, shopping category insights
Naver News MCP	mcp/naver_news_mcp	Korean news search via Naver News API. Sort by relevance/date. Requires Naver Open API auth
PostgreSQL MCP	mcp/postgresql_mcp	Direct read-only connection to a PostgreSQL DB. Enter host, port, user, password, DB name
Database Loader	mcp/DatabaseLoader	Load a pre-configured DB connection. AI auto-uses list_tables / get_schema / query tools
Product Search MCP	mcp/product_search_mcp	Product search with filters — popular, upcoming broadcasts, past broadcasts, currently selling
Slack MCP	mcp/slack_mcp	Connect AI to Slack workspace — send messages, manage channels, search conversations. Slack User Token required
Nano Banana MCP	mcp/nano_banana_mcp	Generate/edit images via Google Gemini. Flash and Pro models, 1K~4K resolution
Atlassian MCP	mcp/atlassian_mcp	Manage Jira issues and Confluence docs via natural language. Cloud and On-premise supported
Microsoft 365 MCP	mcp/ms365_mcp	Connect AI to Microsoft 365 — Outlook Mail, Calendar, Teams, OneDrive, Planner, Excel
API Collection Loader	mcp/APICollectionLoader	Load admin-registered API collection (ToolGraph). AI auto-uses search_tools / call_tool
Web Automation (Playwright)	mcp/WebAutomationMCP	Web-task automation via Playwright browser control. Excel-to-form auto-input with pre-save confirmation
Database Reader	mcp/DatabaseReader	Run fixed SQL queries against a pre-configured DB connection. PostgreSQL, Oracle, Informix
Database Result Processor	mcp/DatabaseResultProcessor	Transform DatabaseReader rows with a full SQL query (sqlite in-memory against table rows) or a no-SQL pipeline (filter, sort, paginate, group-aggregate, column-select). Output as markdown/json/csv/records/scalar

Agent (agents)

Node	ID	Description
Agent Planflow	agents/planflow	Deterministic Plan-and-Execute agent. Intent parsing → graph-based plan → sequential execution → natural-language response
Agent Xgen	agents/xgen	The core AI brain of a workflow. Auto tool selection. OpenAI, Anthropic, Google, AWS models
Agent Harness	agents/harness	Run a saved Harness workflow as a single agent step (system_prompt, tools, strategies, RAG, DB, MCP)

API Loader (api_loader)

Node	ID	Description
API Calling Tool	api_loader/APICallingTool	Build a custom REST API tool and connect it to the Agent. Response-data filtering supported
API Tool Loader	api_loader/APIToolLoader	Load admin-registered API tools from a dropdown instantly

Document Loader (document_loaders)

Node	ID	Description
Search Context	document_loaders/VectorDBContext	Unified document search. Select search mode to configure vector-DB retrieval. Connect to Agent RAG Context
Ontology Search	document_loaders/OntologySearch	Graph-based ontology search using SPARQL + SCS context. Returns relevant triples and source chunks
Search Context (260517)	document_loaders/VectorDBContextV2	Search Context redesign (2026-05-17). Only essential options exposed, advanced tuning uses best-practice defaults

File System (file_system)

Node	ID	Description
My File Storage (Skill)	file_system/filesystem_storage_skill	Grant file-system access as a Skill (SKILL). Bundles 37 fs_* operations behind a single tool, cutting per-call tool-description overhead. Browse, read, create, modify files
Document Adapter	file_system/document_adapter	Edit form documents (DOCX/PPTX/HWPX). 9 tools (inspect_document / get_cell / get_shapes / render_template …)

Memory (memory)

Node	ID	Description
DB Memory (Smart)	memory/db_memory_v3	The smartest conversation memory. Filters unreliable AI responses, time-based decay, smart selection of related past chats

Router (router)

Node	ID	Description
Router	router/Router	Branch data into different paths by key value. Output handles dynamically generated per key value

Tool (tools)

Node	ID	Description
Certificate PDF Tool	tools/Certificate PDF Tool	Auto-generate certificate PDFs from provided data — certificates, awards, completion forms
FloUI v1 (Skill)	tools/floui_v1_skill	FloUI v1 skill node. Bundles the xgen-frontend API catalog (workflow·chat·retrieval·storage·tools·prompt·aichat) plus extended A2UI components into a single SKILL
Show Value	tools/show_any	A pass-through node that displays the value flowing through it on the canvas card. Quickly inspect intermediate values without a separate end node
Tool Output Formatter	tools/output_formatter	Post-processing node that reshapes a tool's response before it reaches the agent. Connect to a Tool node's Output Formatter input port
Workspace Dev (chat→preview)	tools/workspace_dev	Give the agent a persistent dev workspace — write files, run commands, expose a dev-server preview URL. Connect to the Agent's tools input
Hierarchy Tools	tools/hierarchy_tools	Manager-worker Agent hierarchy. Manager delegates subtasks to specialist workers + combines results
Image Loader	image_loader	Load images (URL or upload) into Agent images input for visual analysis
Input Files	input_files	Receive user file uploads. Workflow starting point for processing documents, spreadsheets, images
Input Template	input_template	Dynamic prompts using {{ variable }} placeholders. Reusable prompt patterns
Local CLI Tool	tools/local_cli_tool	Run pre-approved CLI commands on the local machine (Tauri desktop only). Git, Node.js, Python safe execution
Agent Planner	tools/agent_planner	Step-by-step work plan generation. Decompose complex tasks → connect plan output to Agent plan input
Sandbox Code Executor	tools/sandbox_exec	Run code in an isolated throwaway KVM VM. Calculation, data transformation, logic validation
Schema Provider (Input)	input_schema_provider	Define input JSON schema for the AI. Connect to nodes that need structured input
Schema Provider (Output)	output_schema_provider	Define output JSON schema. AI structures responses accordingly for consistent, parseable answers
Value Processor	tools/value_processor	Extract/transform values from structured input (JSON, XML, YAML, CSV, text, regex)
Workbench Prompt	tools/workbench_prompt	Pull centrally-managed, versioned prompts from Workbench Prompt Studio (dev/stg/prd stages)
Workflow Tool	tools/workflow_tool	Use another saved workflow as a tool. Agent can call sub-workflows — modular workflow design

Start Node (startnode)

Node	ID	Description
Input String	input_string	Receive user text input or set a fixed text value. Starting point for text-input workflows

End Node (endnode)

Node	ID	Description
Print Agent Output	tools/print_agent_output	Display Agent output in the workflow UI. Connect to Agent output to surface responses
Print Format	tools/print_format	Custom-templated formatted output. Scores, timestamps, iteration details, to-do lists in structured layout
Send Email	tools/send_email	Give AI email-sending capability. Configure SMTP → Agent composes and sends emails as part of the workflow

Start / End Node Detailed Spec

The start node and end nodes that handle a workflow's I/O have a fairly fixed port/parameter layout, so their detailed specs are listed below. The Item column distinguishes Input (data coming into the node), Output (data the node emits), and Parameter (values set directly in the node's settings panel). Parameters marked "Optional" fall back to their default behavior when left blank.

Input String (input_string)

The starting point of a workflow — receive text input from the user or set a fixed text value. It works in two modes: the user types a question directly, or a predefined text is passed through.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Input format	InputSchema	—	The input-schema definition passed when the node runs. Forwarded automatically from connected nodes.
Output	Text	STR	—	Emits the entered text as a string (STR). Can then be connected to agent, transform, and other nodes.
Parameter	Input value	STR	Optional	Enter text to fix in advance. Leave blank to let the user type it at runtime.
Parameter	Use voice input	BOOL	Optional	Whether to enable speech-to-text (STT). true → a voice-file attach button appears in the prompt UI. false → voice input disabled, text input only.

Print Agent Output (tools/print_agent_output)

Displays an agent's output on the workflow UI screen. Connect it to an agent node's output port to show the final response directly to the user — an end_node.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Output	STREAM | STR	Required / Multiple	Receives output from an agent or text node. The STREAM type handles real-time streaming output; STR handles a completed string.
Output	—	—	—	No output. A display-only end node.
Parameter	—	—	—	No parameters.

Print Format (tools/print_format)

Renders data on screen in a readable format using a customizable template. An end_node that displays scores, timestamps, iteration counts, to-do lists, and the like in a structured layout.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Output	FeedbackDICT	Required	Feedback data in dictionary (DICT) form. Must be structured data containing scores, timestamps, iteration info, TODO items, etc.
Output	—	—	—	No output. A display-only end node.
Parameter	Use formatted output	BOOL	Optional	Whether to use a styled template. true → formatting settings such as Format Style apply. false → emits raw data as-is.
Parameter	Output style	STR	Optional	Choose the output style. default → plain text, cards → card UI, timeline → chronological list, summary only → compact single line.
Parameter	Show score	BOOL	Optional	Whether to show the result's relevance score. true → score next to each result / false → hide score.
Parameter	Show time	BOOL	Optional	Whether to show timestamps. true → creation time next to each result / false → hide timestamp.
Parameter	Max iterations shown	INT	Optional	Set the maximum number of iterations to display on screen, as an integer.
Parameter	Show to-do details	BOOL	Optional	Whether to show Todo/Task details. true → expand details / false → summary only.

Send Email (tools/send_email)

The agent composes and sends email automatically within the workflow. Once SMTP settings are configured, the AI handles the rest — an end_node used for notifications, reports, and similar deliveries.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Content	ANY	Required	The content sent as the email body. The ANY type allows various formats — text, dictionary, agent output, etc.
Output	—	—	—	No output. An email-sending-only end node.
Parameter	SMTP server address	STR	Required	SMTP server address. e.g., smtp.gmail.com, smtp.naver.com
Parameter	SMTP port	INT	Required	SMTP server port number. e.g., 587 (TLS), 465 (SSL), 25 (unencrypted)
Parameter	Sender email (ID)	STR	Required	The sender's email account address. e.g., yourname@gmail.com
Parameter	Sender password	STR	Required	The sender account password or an app-specific password. For Gmail, an App Password is recommended.
Parameter	Recipient email	STR	Required	Recipient email address(es). Separate multiple addresses with a comma (,) or semicolon (;). e.g., a@x.com, b@x.com; c@x.com
Parameter	Include full output	BOOL	Optional	Whether to include the workflow's full output data in the email. true → full output data is appended to the body. false → send only the Content input.

Tool Node Detailed Spec

Nodes in the Tool (tools) category give the agent a specific capability (file generation, image analysis, CLI execution, value transformation, etc.) or assist with workflow I/O. Detailed port/parameter specs for the commonly-used nodes are listed below. The notation matches Start / End Node Detailed Spec; a blank (—) Required cell means the field appears conditionally based on the selected input type or operation.

Certificate PDF Tool (tools/Certificate PDF Tool)

Automatically generates certificate PDFs — completion certificates, awards, diplomas, etc. — from the provided data. The AI produces a fully-formatted certificate file.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the generated PDF tool object. Connect it to an agent node's Tools input.
Parameter	Company seal path	STR	Optional	Path to the company seal (stamp) image file. Used to insert the seal into the certificate.
Parameter	Additional parameters	STR	Optional	Extra parameter values passed when the node runs.

Hierarchy Tools (tools/hierarchy_tools)

Builds a manager-worker agent hierarchy. The manager agent delegates subtasks to specialist worker agents and combines their results. Use it for complex tasks where multiple specialist AIs must collaborate.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Tools	TOOL	Multiple	Connect (multiple) tool nodes to provide to the worker agents below.
Output	Tool	TOOL	—	Emits the assembled hierarchy tool object. Connect it to the manager agent's Tools input.
Parameter	Tool name	STR	Required	The identifier name for this hierarchy tool. Used by the AI to distinguish tools.
Parameter	Tool description	STR	Required	Describe when this tool should be used. The AI references this to decide when to call it.
Parameter	Additional parameters	STR	Optional	Extra parameter values passed when the node runs.

Image Loader (image_loader)

Loads and extracts images for the AI to analyze. Receives images via URL or file upload and passes them to the agent's images input for visual analysis and understanding.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Data source	STR	Required	Receives an image URL or file path as a string.
Output	Images	LIST	—	Emits the loaded images as a LIST. Connect to the agent's images input.
Parameter	Image key name	STR	Optional	The key name to extract the image URL/file path from the data. Default: img_url
Parameter	Allow multiple images	BOOL	Optional	Whether to process multiple images. true → process several images as a list. false → process a single image only.
Parameter	Additional parameters	STR	Optional	Extra parameter values passed when the node runs.

Input Files (input_files)

A workflow start node that receives file uploads from the user. Use it for workflows that process user-uploaded files such as documents, spreadsheets, and images.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	File	FILE	—	Emits the uploaded file as FILE. Connect to the next node that processes the file.
Parameter	File	STR	Optional	The field where the user uploads a file at runtime. You may also enter a pre-fixed file path.

Input Template (input_template)

Generates dynamic prompts using {{ variable }} placeholders. Once you author a template, placeholders are replaced with actual values at runtime. Ideal for building reusable prompt patterns.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Variables	STR	Multiple	Receives the values (STR) that fill the template's {{ variable }} slots. Multiple variables can be connected.
Output	Prompt	STR	—	Emits the completed prompt string with variables substituted.
Parameter	Rule input method	STR	Optional	Choose the template input method. Direct input → write directly in the Template field below; Select template → pick from the saved template list.
Parameter	Rule template	STR	Optional	Select from the saved template list. Only User Prompt type prompts are shown.
Parameter	Template content	STR	Required	Author the prompt template directly. Mark dynamic parts as {{ variable_name }}.
Parameter	Required-input check	BOOL	Optional	Whether to error on a missing template variable. true → raise an error if a variable is missing. false → treat missing variables as empty values.

Local CLI Tool (tools/local_cli_tool)

Lets the AI run pre-approved CLI commands on the local machine (Tauri desktop app only). Define the allowed commands in Skills and the AI can safely run Git, Node.js, Python commands, and more. For security, only commands defined in Skills can be executed.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the CLI-execution tool object. Connect to the agent's Tools input to grant CLI-execution capability.
Parameter	Skill definition (JSON)	STR	Required	Define the allowed CLI commands as a JSON array. Key fields: • id: unique identifier (alphanumeric, underscore) • name: tool display name • description: tool description (used by the AI to decide when to call) • command: base command to run • allowed_args: list of allowed arguments (incl. type, description, required) • timeout: execution time limit (seconds) • requires_confirmation: if true, ask the user to confirm before running • risk_level: safe / moderate / dangerous
Parameter	Skill preset	STR	Optional	Selecting a predefined Skill Preset auto-fills the Skills JSON.

Agent Planner (tools/agent_planner)

Generates a step-by-step work plan for the agent to follow. The AI decomposes a complex task into manageable steps and executes them in order. Connect the Plan output to the agent's plan input.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Plan	PLAN	Optional	Receives plan data from a previous step and continues from it. An optional connection.
Input	Tools	TOOL	Multiple	Connect (multiple) tool nodes the agent will use during plan execution.
Output	Plan	PLAN	—	Emits the generated step-by-step plan as PLAN. Connect to the agent's plan input.
Parameter	Plan description	STR	Optional	Describe the work plan for the agent to execute, in text. State the plan's goal and steps.

Schema Provider (Input) (input_schema_provider)

Defines the input format for the AI to follow. Generate a JSON schema specifying the expected parameters, then connect it to a node that needs structured input.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Input format	InputSchema	—	Emits the defined input schema. Connect to a node that needs schema input, such as a user-input node.
Parameter	kwargs (enter key names directly)	STR	Optional	Define the parameters the AI receives. Enter a key name and select a type. Type options: STRING / INTEGER / FLOAT / BOOLEAN / OBJECT / LIST

Schema Provider (Output) (output_schema_provider)

Defines the output format of the AI's response. Generate a JSON schema and the AI returns a structured answer that conforms to it. Use it when you need consistent, parseable responses.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Output format	OutputSchema	—	Emits the defined output schema. Connect to an agent node's output-schema input.
Parameter	kwargs (enter key names directly)	STR	Optional	Define the parameters the AI receives. Enter a key name and select a type. Type options: STRING / INTEGER / FLOAT / BOOLEAN / OBJECT / LIST

Value Processor (tools/value_processor)

Extracts or transforms values from structured input (JSON, XML, YAML, CSV, Text, Regex). Specify the input type, operation, and path/key to pull out the value you want. Handles both Python objects and string forms.

Selecting the Expected input type first exposes only the relevant parameters. The table below is the full list aggregating parameters across all input types.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Input	ANY	Required	Receives the data to process. Supports various formats — JSON, XML, YAML, CSV, text, regex, etc.
Output	Output	ANY	—	Emits the extracted/transformed result.
Parameter	Expected input type	STR	Required	Specify the input data format. All subsequent options branch on this setting. Options: JSON / XML / YAML / CSV / Text / Regex
Parameter	List input	BOOL	—	Enable when the input is a list/array. true → apply the operation to each item and return a list.
Parameter	JSON operation	STR	—	The operation to perform on a JSON/dictionary input.
Parameter	Path	STR	—	The extraction path in dot/bracket notation. e.g., user.name / items[0].title / items[*].id
Parameter	Paths	STR	—	Multiple comma-separated paths. Rename keys with the alias=path form.
Parameter	Keys	STR	—	Top-level keys to keep, comma-separated.
Parameter	Keys to Omit	STR	—	Top-level keys to remove, comma-separated.
Parameter	Default Value	STR	Optional	The default returned when a path is missing. Auto-parsed as number/bool/JSON.
Parameter	Strict (raise on missing)	BOOL	Optional	Whether to error when a path/key is missing. true → raise an error / false → return the default.
Parameter	XML Operation	STR	—	The operation to perform on an XML input.
Parameter	Tag Name	STR	—	The tag name to find. Use the {uri}local form with namespaces.
Parameter	Tag@Attribute	STR	—	Extract an attribute value with the tag@attribute_name form. *@attr reads from the root.
Parameter	XPath	STR	—	Select elements with ElementTree-compatible XPath. e.g., .//book[@id='1']/title
Parameter	Find All Matches	BOOL	Optional	true → return a list of all matches / false → return the first match only.
Parameter	Text Only	BOOL	Optional	true → return the inner text of matched elements / false → return the element XML string.
Parameter	Namespaces (JSON)	STR	Optional	A namespace JSON map for XPath. e.g., {"ns": "http://example.com"}
Parameter	Path (for XML)	STR	—	Dot-notation path, same as XML. e.g., a.b[0].c / items[*].name
Parameter	Default Value (for XML)	STR	Optional	The default returned when a path is missing.
Parameter	Strict (for CSV)	BOOL	Optional	Whether to error when a path is missing.
Parameter	Has Header	BOOL	—	true → treat the first row as column names.
Parameter	Delimiter	STR	—	The field delimiter. Use \t for tab-separated.
Parameter	Column	STR	—	The column name to extract (when a header exists) or a 0-based index. Leave blank to return the whole row.
Parameter	Row Index	INT	Optional	The data row number to extract (0-based, excluding the header). -1 returns all rows.
Parameter	Text Operation	STR	—	The operation to perform on a text input.
Parameter	Separator	STR	—	The delimiter string used when splitting.
Parameter	Index	INT	Optional	The item index to extract after splitting lines (0-based, negatives allowed). -1 returns all.
Parameter	Slice Start	INT	—	The slice start index.
Parameter	Slice End	INT	—	The slice end index (exclusive).
Parameter	Substring	STR	—	The substring to check for. Returns a BOOL value.
Parameter	Pattern	STR	—	A Python regex pattern. Using groups (...) extracts the group contents.
Parameter	Group	INT	—	The capture-group index. 0 returns the whole match.
Parameter	Find All	BOOL	Optional	true → return a list of all matches / false → return the first match only.
Parameter	Flags	STR	Optional	Combine regex flags. i → ignore case / m → multiline / s → dot (.) matches newline / x → verbose mode.
Parameter	Output Format	STR	Optional	The final output format. auto → return as native Python types / json_string → always serialize to a JSON string.
Parameter	Wrap In Key	STR	Optional	If not empty, wraps the result in a {key: value} dictionary.

Workflow Tool (tools/workflow_tool)

Calls another saved workflow as a tool from within the current workflow. The agent can run sub-workflows on demand, enabling modular workflow design.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Input data	ANY	—	Receives the input data to pass to the sub-workflow.
Output	Result	STR	—	Emits the sub-workflow's execution result as a string (STR).
Parameter	Workflow to run	STR	Required	Enter the name of the workflow to run.
Parameter	Stream in real time	BOOL	Optional	Whether to stream in real time. true → forward the real-time stream to the next node (returns a Generator). false → collect all results, then return a string.

Agent Node Detailed Spec

Nodes in the Agent (agents) category act as the core AI brain of a workflow. Connect tools, documents, memory, and the like, and the agent uses them to answer user questions. The notation matches Start / End Node Detailed Spec.

Agent Planflow (agents/planflow)

A deterministic Plan-and-Execute agent for a single API collection. The AI parses intent, builds a plan on a graph, then executes it in order (no per-step LLM). Faster and easier to audit than a ReAct agent in API orchestration.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Text	STREAM STR | STR	Required	Receives the user question or instruction. Supports both STREAM and STR types.
Input	Prior Entities	DICT | STR	—	Receives entity information extracted in a previous step.
Output	Stream	STREAM STR	Stream	Emits the response as a real-time stream.
Output	Result	STR	—	Emits the completed response as a string.
Parameter	API Collection	STR	Required	Select an API collection registered in Admin.
Parameter	Provider	STR	Required	The LLM provider used for Stage 1 (intent parsing) and Stage 4 (response generation).
Parameter	OpenAI Model	STR	Required	The OpenAI model name to use.
Parameter	Anthropic Model	STR	Required	The Anthropic model name to use (via litellm).
Parameter	Streaming	BOOL	—	Deliver results to the client as a real-time stream (recommended).
Parameter	Top K	INT	Optional	Stage 1 catalog size — the number of candidate tools to search.
Parameter	Auth Token Override	STR	Optional	Override the Bearer token. Leave blank to use the collection's AuthProfile.
Parameter	API Base URL Override	STR	Optional	Override the collection's base_url (optional).

Agent Xgen (agents/xgen)

The core AI brain node of a workflow. Connect various tools — DB queries, document search, web search, etc. — and the agent auto-selects them to answer user questions. Supports OpenAI, Anthropic, Google, AWS Bedrock, and vLLM providers.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Text	STREAM STR | STR	Multiple	User question or instruction. Supports STREAM and STR types; multiple connections allowed.
Input	Files	FILE	Multiple	Files for the agent to analyze. Connect with the Input Files node.
Input	Images	LIST	Multiple	Image list for the agent to analyze. Connect with the Image Loader node.
Input	Tools	TOOL	Multiple	Tools the agent uses. Connect (multiple) with Tool-family nodes.
Input	Skills	SKILL	Multiple	Skills the agent uses. Connect with Skill-family nodes (e.g., FileSystem Skill). Each skill bundles many operations behind a single tool, cutting per-call context overhead.
Input	Memory	OBJECT	—	Conversation memory object. Connect with the DB Memory node to let the AI remember previous conversations and give more coherent replies using past context.
Input	Search Context	DocsContext	Multiple	Document-search context. Connect with the Search Context node.
Input	Output format	OutputSchema	—	Output schema definition. Connect with the Schema Provider (Output) node.
Input	Plan	PLAN	—	Work plan. Connect with the Agent Planner node.
Output	Stream	STREAM STR	Stream	Emits the response as a real-time stream — text appears character by character as the AI generates it.
Output	Result	STR	—	Complete response text. The full answer is returned after the AI finishes generating.
Parameter	AI engine	STR	Required	Select the AI provider: openai (ChatGPT), anthropic (Claude), google (Gemini), bedrock (AWS AI), vllm (self-hosted).
Parameter	OpenAI model	STR	Optional	The OpenAI model to use. Applies when Provider is openai. e.g., gpt-4.1, gpt-4o, gpt-4.1-mini
Parameter	Anthropic model	STR	Optional	The Anthropic model to use. Applies when Provider is anthropic.
Parameter	Google model	STR	Optional	The Google model to use. Applies when Provider is google.
Parameter	AWS Bedrock model	STR	Optional	The AWS Bedrock model to use. Applies when Provider is bedrock.
Parameter	vLLM model	STR	Optional	The vLLM model to use. Applies when Provider is vllm.
Parameter	Answer creativity	FLOAT	Optional	Adjust the AI creativity level. Range: 0~2. Closer to 0 = consistent, factual answers; closer to 2 = diverse, creative answers.
Parameter	Max answer length (max tokens)	INT	Optional	Maximum length of the AI response (tokens). Range: 1~65536
Parameter	Tool-call count	INT	Optional	Max number of tool calls the AI can make in one turn. Range: 1~100. Increase for complex multi-step tasks.
Parameter	File processing mode	STR	Optional	Choose how uploaded files are processed. For scanned documents, enhanced_ocr is recommended.
Parameter	Max images processed	INT	Optional	Max number of images the AI can process at once. Range: 0~100
Parameter	Always show sources	BOOL	Optional	Whether to always show sources when referencing documents.
Parameter	Prompt Source	STR	—	Choose how the system prompt is provided. direct → write directly in System Prompt below; template → pick from saved prompt templates.
Parameter	Template	STR	—	Select a saved prompt template. Only System Prompt type is shown. Enabled when prompt_source is template; content is editable after selection.
Parameter	System Prompt	STR	—	Write the AI's role and instructions. e.g., You are a friendly financial advisor. Answer loan and deposit questions accurately.
Parameter	Show intermediate steps	BOOL	Optional	Output the AI's reasoning process (tool calls, intermediate results, reasoning steps). Useful for debugging.
Parameter	Apply safety filter	BOOL	Optional	Enable a safety filter that blocks inappropriate content (profanity, harmful info). Recommended for production.
Parameter	Show AI typing	BOOL	Optional	Enable real-time chat-style output. true → text streams character by character. false → output the completed answer all at once.
Parameter	Show result immediately (multi-agent)	BOOL	Optional	Whether to display this agent's response on screen when multiple agents are connected. false → process internally and pass to the next agent.
Parameter	Answer separately	BOOL	Optional	Whether to display each agent's response in a separate area when multiple agents are connected.

Agent Harness (agents/harness)

Runs a saved Harness workflow as a single agent step. Uses all of the selected workflow's stage settings (system_prompt, selected tools, strategy, RAG/DB/MCP connections, etc.) as-is. The node acts as an input wrapper (one MCP).

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Text	STR	Required	The text input to pass to the Harness workflow.
Output	Response	STR	—	Emits the Harness workflow's execution result as a string.
Parameter	Harness workflow	STR	Required	Select the workflow to run from the saved Harness workflow list. The workflow's saved stage settings apply as-is.

API Loader Node Detailed Spec

Nodes in the API Loader (api_loader) category turn an external REST API into a tool the agent can call.

API Calling Tool (api_loader/APICallingTool)

Builds a custom API tool and connects it to the agent. Once you configure the REST API endpoint, the agent calls it automatically when needed during a conversation. Response-data filtering lets you extract only the data you need.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Input format	InputSchema	—	The input schema passed on API call. Connect with the Schema Provider (Input) node to define a structured input format.
Output	Tool	TOOL	—	Emits the generated API tool object. Connect to the agent's Tools input.
Parameter	Tool name	STR	Required	The tool's unique name. The agent identifies and calls the tool by this name.
Parameter	Tool description	STR	Required	Describe to the AI when this tool should be used. e.g., Use this when asked about interest rates or exchange rates.
Parameter	API address	STR	Required	Enter the API URL to call. e.g., https://api.example.com/rates
Parameter	HTTP method	STR	Required	Select the HTTP method: GET, POST, PUT, DELETE, PATCH
Parameter	Response timeout (sec)	INT	—	API response timeout (seconds). Range: 1~300
Parameter	Use response filtering	BOOL	—	Whether to enable response filtering that extracts only specific data. true → the Response Filter Path / Fields settings below apply.
Parameter	Response filter path	STR	—	Specify the JSON path of the data to extract. e.g., payload.searchDataList
Parameter	Response filter fields	STR	—	Enter field names to extract, comma-separated. e.g., interestRate,productNm

API Tool Loader (api_loader/APIToolLoader)

Quickly loads an API tool pre-registered in the admin panel. Instead of configuring API settings yourself, just pick an already-configured tool from the dropdown.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the loaded API tool object. Connect to the agent's Tools input.
Parameter	Select API tool	STR	Required	Select the tool to use from the list of registered API tools.

Document Loader Node Detailed Spec

Nodes in the Document Loader (document_loaders) category search documents from a vector DB or knowledge graph and provide them as the agent's RAG context.

Search Context (document_loaders/VectorDBContext)

A unified document-search node. Select a search mode to configure how documents are retrieved from the vector database. Connect to the agent's RAG Context input.

Advanced tuning parameters

Beyond the core parameters below, this node exposes ~40 advanced tuning parameters (hybrid search, multi-stage search, ontology graph, reranking, filtering LLM, etc.). Defaults are fine for most cases, so only the core items are listed below. For most users we recommend the Search Context (260517) node (document_loaders/VectorDBContextV2), which exposes only the essential options and handles the rest with best-practice defaults.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Reference docs	DocsContext	—	Emits the retrieved document context. Connect to the agent's RAG Context input.
Parameter	Search mode	STR	Required	Select the search mode. Light / Light+ / Hard → created as a tool the agent can call; Always Search → runs the search automatically at prep time.
Parameter	Collection name	STR	Required	Select the vector-DB collection to search.
Parameter	Tool description	STR	Required	Describe when this tool should be used. Used by the AI to decide when to call it.
Parameter	Top results	INT	Optional	The number of top results returned by vector search (applies in Light/Light+/Always Search modes).
Parameter	Min relevance score	FLOAT	Optional	The minimum similarity score for inclusion. Range: 0.0~1.0
Parameter	Use reranking	BOOL	Optional	Whether to improve result accuracy with cross-encoder reranking.
Parameter	Rerank candidates	INT	Optional	The number of top candidates to rerank.
Parameter	Search-augmentation prompt	STR	Optional	A prompt to improve responses using the RAG context.
Parameter	Tool name	STR	Optional	The name identifier of the search tool (ignored in Always Search mode).
Parameter	Strict source display	BOOL	Optional	Whether to force source display when referencing documents.

Ontology Search (document_loaders/OntologySearch)

A graph-based ontology-search node using SPARQL and SCS context. Queries a pre-built knowledge graph to find relevant triples and source chunks for a structured answer.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the ontology-search tool as TOOL. Connect to the agent's Tools input.
Parameter	Tool name	STR	Required	The name identifier of this tool.
Parameter	Tool description	STR	Required	Describe when the agent should use this tool.
Parameter	Collection name	STR	Required	Select the collection where the ontology is built.
Parameter	Use hierarchy context	BOOL	Optional	Whether to enable the SCS context profile for hierarchy-aware answers.
Parameter	Max source chunks	INT	Optional	The maximum number of source chunks to include in the answer.
Parameter	Multi-turn search	BOOL	Optional	Whether to enable multi-turn ReAct graph traversal for complex multi-hop questions.
Parameter	Max search turns	INT	Optional	The maximum number of search turns in multi-turn mode.

File System Node Detailed Spec

Nodes in the File System (file_system) category grant the AI file-storage access and document-editing abilities.

My File Storage (Skill) (file_system/filesystem_storage_skill)

Grants the AI file-system access as a Skill (SKILL). Bundles 37 fs_* operations (browse, read, create, modify, etc.) behind a single tool, greatly reducing per-call tool-description overhead versus TOOL mode. On first use the agent fetches the catalog once via action='help'.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Skill	SKILL	—	Emits the assembled file-storage skill. Connect to the agent's Skills (SKILL) input.
Parameter	Storage folder	STR	Required	Select the storage folder to use. The path is relative to the file-storage/{user_id}/ subfolder.
Parameter	Max rows	INT	Optional	Max rows returned in a single read.
Parameter	Max column width	INT	Optional	Max character width per column when displaying tabular data. Excess is truncated with ....
Parameter	Max chars per page	INT	Optional	Max characters shown per page.
Parameter	Max lines per read	INT	Optional	Max lines returned in a single read.

Memory / Router Node Detailed Spec

The Memory (memory) category handles conversation memory, and the Router (router) category handles data-flow branching.

DB Memory (Smart) (memory/db_memory_v3)

The smartest conversation-memory node. Beyond simple recall, it filters low-confidence AI responses, automatically decays the weight of older information over time, and intelligently selects the most relevant past conversations.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Current input	STR	—	Receives the current conversation input, stores it in memory, and searches for related context.
Output	Memory	OBJECT	—	Emits the stored memory object. Connect to the agent's memory input.
Output	Context	STR	—	Emits past-conversation context related to the current input as a string.
Parameter	Conversation session ID	STR	—	Set a unique ID to distinguish conversation sessions. The same ID groups into the same conversation.
Parameter	Max conversations	INT	Optional	Max messages to retain. 0 retains all. Range: 0~100
Parameter	Max characters	INT	Optional	Set the maximum character count of retained conversations.
Parameter	Recent messages remembered	INT	Optional	The number of recent messages passed directly to the AI. Range: 2~10
Parameter	Top-K similar sources	INT	Optional	The number of top related messages to include. Range: 0~20
Parameter	Confidence threshold	FLOAT	Optional	The minimum confidence score for including a message. Range: 0~1
Parameter	Delete old conversations	BOOL	Optional	Whether to enable time-based decay. true → older messages gradually lose importance.
Parameter	Include AI reasoning memory	BOOL	Optional	Whether to include the AI's reasoning process in memory. true → AI thinking content is also stored.

Router (router/Router)

Routes data to different paths based on a key value. Set the routing criteria (e.g., language, category) and an output path is dynamically created for each key value, with input data flowing into the matching path.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	Input data	ANY	Required	Receives the data to route. The ANY type accepts any format.
Output	—	—	—	Output handles are dynamically created per the Routing Criteria setting. A separate output port is made for each key value.
Parameter	Routing Criteria	STR	Required	Enter the key name to route on. The router checks this key's value in the input data and forwards to the matching output path. e.g., language, category, action

MCP Node Detailed Spec

Nodes in the MCP (mcp) category turn external MCP servers and APIs into tools (TOOL) the agent can use. Most have no Input; you specify the connection target via parameters, then connect the output tool to the agent's Tools input — a common pattern across the category. The notation matches Start / End Node Detailed Spec.

MCP Tool Loader (mcp/MCPLoader)

Connects to an MCP server and loads all its tools at once. Instead of configuring individual MCP nodes, this single node gives access to all tools on the MCP server. Just select an active MCP session.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the loaded MCP tool object. Connect to the agent's Tools input.
Parameter	MCP session	STR	Required	Select the active MCP session to load tools from.

Tavily Search MCP (mcp/tavily_search_mcp)

Searches the web with Tavily, an AI-optimized search engine. Supports structured results, domain filtering, summary answers, raw content extraction, and search-depth settings. Suited for AI research tasks.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the Tavily search tool object. Connect to the agent's Tools input.
Parameter	Tavily API key	STR	Required	Enter the Tavily API key.
Parameter	Max results	INT	Optional	Set the maximum number of search results to return.
Parameter	Tool description	STR	Optional	Describe to the AI when to use this search tool. Leave blank to use the default behavior.
Parameter	Include domains	STR	Optional	Domains to include in results, comma-separated.
Parameter	Exclude domains	STR	Optional	Domains to exclude from results, comma-separated.
Parameter	Include answer	BOOL	Optional	Provide a short summary answer alongside results. true → include summary answer.
Parameter	Include raw content	BOOL	Optional	Include each result's raw HTML content. true → include raw HTML.
Parameter	Include images	BOOL	Optional	Include relevant images in results. true → include images.
Parameter	Search depth	STR	Optional	Search depth. basic → fast results, advanced → more comprehensive results.

Brave Search MCP (mcp/brave_search_mcp)

Searches the web in real time using the Brave Search API. Connect it and the AI can search the internet for up-to-date info. Supports country, period, and result-count filtering; a Brave API key is required.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the Brave search tool object. Connect to the agent's Tools input.
Parameter	Brave API key	STR	Required	Enter the Brave Search API key. Available from brave.com.
Parameter	Result count	INT	Optional	Set the maximum number of search results to return.
Parameter	Country	STR	Optional	Result country filter. e.g., kr=Korea, us=USA, jp=Japan
Parameter	Period filter	SELECT	Optional	Time-range filter. Default → all, pd (Past Day) → last day, pw (Past Week) → last week, pm (Past Month) → last month, py (Past Year) → last year.
Parameter	Text highlight	BOOL	Optional	Include text emphasis (bold/highlight) in results. true → include highlighting.

EPG DAUM MCP (mcp/epg_daum_mcp)

Fetches Korean home-shopping TV schedules from DAUM. The AI can look up broadcast times, program names, and channel info. Supports caching for repeated queries.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the DAUM EPG tool object. Connect to the agent's Tools input.
Parameter	Use Cache	BOOL	Optional	Whether to use cached data. true → use cached data for a fast response, false → fetch fresh data.

EPG NAVER MCP (mcp/epg_naver_mcp)

Fetches Korean home-shopping TV schedules from NAVER. Provides the same features as EPG DAUM MCP, using NAVER as the data source.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the NAVER EPG tool object. Connect to the agent's Tools input.
Parameter	Use Cache	BOOL	Optional	Whether to use cached data. true → use cached data for a fast response, false → fetch fresh data.

GitHub MCP (mcp/github_mcp)

Manages GitHub repositories via natural language. The AI can handle repo, issue, pull-request, and code search. Uses GitHub App authentication (App ID + Private Key).

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the GitHub management tool object. Connect to the agent's Tools input.
Parameter	GitHub App ID	STR	Required	Enter the GitHub App ID.
Parameter	GitHub App Private Key	STR	Required	Enter the GitHub App Private Key in PEM format.
Parameter	GitHub Repository	STR	Required	Enter the target repository. Format: owner/repo

GitLab MCP (mcp/gitlab_mcp)

Manages GitLab projects via natural language. The AI can search projects, manage branches, edit files, handle issues, and create Merge Requests. Supports both GitLab.com and self-hosted GitLab instances.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the GitLab management tool object. Connect to the agent's Tools input.
Parameter	GitLab URL	STR	Optional	Enter the GitLab instance URL. Default: https://gitlab.com (a self-hosted URL also works).
Parameter	GitLab Personal Access Token	STR	Required	Enter the GitLab Personal Access Token. Issue at: Settings > Access Tokens.

Meta Search MCP (mcp/meta_search_mcp)

The AI automatically finds and crawls relevant websites to gather comprehensive info. No API key needed; the AI decides which sites to search and aggregates results across them.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the meta-search tool object. Connect to the agent's Tools input.
Parameter	Max results	INT	Optional	Set the maximum number of results to return.

Naver Datalab MCP (mcp/naver_datalab_mcp)

Accesses Naver Datalab for search-trend analysis and shopping insights. The AI can look up popular search terms, search-volume trends, and shopping-category insights. Useful for market research and trend analysis.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the Naver Datalab tool object. Connect to the agent's Tools input.
Parameter	Description	STR	Required	Describe to the AI when to use this analytics tool.

Naver News MCP (mcp/naver_news_mcp)

Searches Korean news via the Naver News API. Connect it and the AI finds the latest Korean news articles to use in answers. Can sort by relevance or date. Requires Naver Open API credentials.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the Naver news search tool object. Connect to the agent's Tools input.
Parameter	Description	STR	Required	Describe to the AI when to use this news search tool.
Parameter	Naver Client ID	STR	Required	Enter the Naver application Client ID.
Parameter	Naver Client Secret	STR	Required	Enter the Naver application Client Secret.
Parameter	Sort	STR	Optional	Result sort order. sim → by relevance, date → by most recent.

PostgreSQL MCP (mcp/postgresql_mcp)

Connects directly to a PostgreSQL database and runs read-only queries. Connect by entering host, port, username, password, and database name. Unlike Database Loader, it connects directly without DB Connection Manager — suited for quick query setups.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the PostgreSQL query tool object. Connect to the agent's Tools input.
Parameter	Host	STR	Required	Enter the PostgreSQL server address. e.g., 192.168.0.10, localhost
Parameter	Port	INT	Required	Enter the port number. Default: 5432
Parameter	Username	STR	Required	Enter the database login username.
Parameter	Password	STR	Required	Enter the database login password.
Parameter	Database	STR	Required	Enter the name of the database to connect to.
Parameter	DB Prompt	STR	Optional	Describe the database so the AI can write better queries. e.g., Loan DB with repayment history.
Parameter	Tool Prefix	STR	Optional	A prefix to prepend to tool names when using multiple DBs at once. e.g., hr → hr_list_tables, hr_query. Leave blank for a single DB.
Parameter	Schema	STR	Optional	Enter the PostgreSQL schema name. Default: public. e.g., app_main
Parameter	Sample Rows	INT	Optional	The number of sample rows per table during schema inspection. 0 returns schema info only.

Database Loader (mcp/DatabaseLoader)

Loads a pre-configured database connection so the AI can query data. Select a connection in DB Connection Manager and the AI auto-uses the list_tables, get_schema, and query tools. Supports PostgreSQL, Oracle, Informix.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the database query tool object. Connect to the agent's Tools input.
Parameter	DB connection	STR	Required	Select a database connection from the DB Connection Manager list.
Parameter	Access password	STR	Required	Enter the password set when creating the DB connection. Required for security verification.
Parameter	DB description	STR	Optional	Describe the database so the AI can write better queries. e.g., Loan DB. The loan_records table holds loan history and repayment data.
Parameter	Tool prefix	STR	Optional	A prefix to prepend to tool names when using multiple DBs at once. e.g., hr → hr_list_tables, hr_get_schema, hr_query. Leave blank for a single DB.
Parameter	Sample rows	INT	Optional	The number of sample rows per table to show during schema inspection. Helps the AI understand the data format. 0 returns schema info only.

Product Search MCP (mcp/product_search_mcp)

Searches product info using various filters. Look up popular products, upcoming broadcasts, past broadcasts, and currently-selling products. You can control the result count and whether images are included.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the product search tool object. Connect to the agent's Tools input.
Parameter	Tool description	STR	Required	Describe to the AI when to use this product search tool.
Parameter	Search type	STR	Optional	Select the search type. popular → popular, future → upcoming broadcasts, past → past broadcasts, sales → currently selling. Leave blank to let the AI decide.
Parameter	Max results	INT	Optional	The maximum number of results to return. Default: 10
Parameter	Include images	BOOL	Optional	Include product images in results. true → include images.

Slack MCP (mcp/slack_mcp)

Connects the AI to a Slack workspace. The AI can send messages, manage channels, and search conversations. A Slack User Token (xoxp-) is required.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the Slack management tool object. Connect to the agent's Tools input.
Parameter	Slack User Token	STR	Required	Enter the Slack User Token. A token starting with xoxp-.

Nano Banana MCP (mcp/nano_banana_mcp)

A Gemini-based image generation and editing tool (Nano Banana). Generate images from text prompts or edit existing images with AI.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the image generation/editing tool object. Connect to the agent's Tools input.
Parameter	Google API Key	STR	Required	Enter the Gemini API key issued from Google AI Studio.
Parameter	Model	SELECT	Required	Select the generation model. Flash → fast generation, Pro → high-quality generation.
Parameter	Image Size	SELECT	Optional	Set the output image resolution. 4K is available only on the Pro model.
Parameter	Response Modality	SELECT	Optional	Select the response format.

Atlassian MCP (mcp/atlassian_mcp)

Manages Jira issues and Confluence documents via natural language. The AI can create issues, manage projects, author wiki pages, and search both platforms. Supports both Cloud and On-premise Atlassian.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the Atlassian management tool object. Connect to the agent's Tools input.
Parameter	Atlassian URL	STR	Required	Enter the Atlassian URL. e.g., https://your-domain.atlassian.net
Parameter	Email	STR	Required	Enter the Atlassian account email.
Parameter	API Token	STR	Required	Enter the Atlassian API Token. Issue at: https://id.atlassian.com/manage/api-tokens
Parameter	Tools Scope	STR	Optional	Select the tool scope. jira → Jira tools only, confluence → Confluence tools only, both → both (default).

Microsoft 365 MCP (mcp/ms365_mcp)

Connects the AI to Microsoft 365. Supports Outlook Mail, Calendar, Teams, OneDrive, Planner, Excel, and more. Use feature presets to load only the tools you need. Initial setup requires Device Code Flow authentication in Admin > MCP Station.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the Microsoft 365 tool object. Connect to the agent's Tools input.
Parameter	Feature preset	STR	—	Select which MS 365 services to enable. e.g., all, mail-only, calendar-only, etc.
Parameter	Organization mode	BOOL	—	Include organizational features such as Teams, SharePoint, Planner. true → enable organizational features.

API Collection Loader (mcp/APICollectionLoader)

Loads a registered API collection (ToolGraph) so the AI can search and call APIs in natural language. Select a collection from the dropdown and the AI auto-uses search_tools and call_tool.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the API collection tool object. Connect to the agent's Tools input.
Parameter	API collection	STR	Required	Select an API collection registered in Admin.
Parameter	Max tools	INT	Optional	The maximum number of tools returned per search query. Default: 5
Parameter	API base URL override	STR	Optional	Override the collection's base_url (optional).
Parameter	Auth token override	STR	Optional	Override the Bearer token. Leave blank to use the collection's AuthProfile (recommended). Enter a value for temporary testing only.
Parameter	Tool prefix	STR	Optional	A prefix to prepend to generated tool names when chaining multiple loaders. e.g., erp → erp_search_tools, erp_call_tool

Web Automation (Playwright) (mcp/WebAutomationMCP)

Automates web tasks via Playwright browser control. Read data from Excel to auto-fill web forms, with user confirmation before saving. Supports login automation, page navigation, and complex form workflows.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Tool	TOOL	—	Emits the web-automation tool object. Connect to the agent's Tools input.
Parameter	Playwright MCP session	STR	Required	Select the Playwright MCP session to use.
Parameter	Target website URL	STR	Required	Enter the target website URL. e.g., https://erp.company.com
Parameter	Login config (JSON)	JSON	Required	Enter the login configuration as JSON. Use environment variables in ${VAR} form.
Parameter	Page config (JSON)	JSON	Required	Set page key-to-path mappings as JSON.
Parameter	Form config (JSON)	JSON	Required	Set form-field mappings and fixed values as JSON.

Database Reader (mcp/DatabaseReader)

Runs a fixed SQL query against a pre-configured database connection and returns the result as text. No AI needed; select a connection, enter a query, and the result flows to the next node. Supports PostgreSQL, Oracle, Informix.

Item	Port / Parameter	Type	Required	Description & Behavior by Value
Input	—	—	—	No input.
Output	Query result	STR	—	Emits the SQL query result as a string (STR).
Output	Rows (structured)	LIST	—	Also emits the result as a structured row list (LIST). Connect to the Database Result Processor node for further processing.
Parameter	DB Connection	STR	Required	Select a database connection from the DB Connection Manager list.
Parameter	Custom Password	STR	Required	Enter the password set when creating the DB connection.
Parameter	SQL Query	STR	Required	Enter the SQL query to run. Only SELECT / WITH queries are allowed.
Parameter	Max Rows	INT	Optional	The maximum number of rows to return. 0 means no limit.

Node Management Screen (Admin)

A system administrator can browse and search the same node catalog from the Admin Center → Agent Operations → Node Management screen (view ID admin-node-management). The page header reads "Node Management — Manage and explore agentflow nodes."

Area	Content
Top controls	Table / Tree view toggle; search box (partial match against category, function, node name, and tag)
Body	Node catalog rendered as a table or a tree depending on the selected view

Browse with the Tree View

1. Switch the top toggle to Tree
2. Expand a category in the tree on the left
3. Click an item — the right panel shows node details (I/O spec, tags, description)

Browse with the Table View

1. Switch the top toggle to Table
2. Sort and filter columns to compare nodes side-by-side
3. Click a row — the right panel shows the same detail view

Search

Typing in the top search box returns partial-match results instantly:

• Category — e.g., Agent, Tool
• Function — e.g., search, invoke
• Node name — e.g., Document Loader
• Tag — labels attached to nodes

Operational Recommendations

• Consistent category naming — When adding custom nodes, reuse existing categories. Split into a new category only after enough nodes accumulate.
• Use tags — Tag each node by function, domain, and owning department to improve search efficiency.
• Monitor deployed Agent nodes — Items under the Agent category mirror the environment's deployed agents. Audit periodically to ensure undeployed / inactive agents do not remain exposed.

Related Chapters

• Creating an Agent · Adding Nodes — Procedure for adding nodes on the canvas
• Agent Operations — The admin Agent Operations group menus (Agent Management, Node Management, etc.)
• MCP Library — Manage MCP nodes that appear under the MCP category

Contact

For inquiries about nodes, contact the Xgen Solution Administrator.