Tool Node 🔧
Extends an AI Task with callable tools for function calling / agent capabilities.
Node type: tool
Category: AI
Actor: (depends on tool subtype)
Description
Tool nodes are attached to an AI Task via toolConnection links. When the AI decides to use a tool (via the LLM's function calling capability), the engine executes the connected Tool node and returns the result to the AI.
This enables the AI Task to act as an intelligent agent that can:
- Search a knowledge base for answers
- Query a database
- Call external REST APIs
- Send emails
- Execute web searches
- Manage files
The AI orchestrates which tools to use, in what order, and with what parameters — you only configure the tools and connect them.
Tool Subtypes
There are 7 tool subtypes, each functioning like its equivalent Service Task but callable by the AI:
| Subtype | Icon | Integration |
|---|---|---|
| 📬 | Send email via mail provider | |
| SQL | 🔍 | Query database |
| REST API | 🔌 | Call external REST API |
| MCP | 🔗 | Call Model Context Protocol server |
| Knowledge Base | 📚 | Search vector store / RAG collection |
| Web Search | 🔍 | Search the web |
| File Manager | 📁 | Read, create, update, or delete files |
Connecting Tools to AI Task
In the workflow designer:
- Add an AI Task node
- Add one or more Tool nodes nearby
- Draw
toolConnectionlinks from the AI Task to each Tool - Configure each Tool's properties
The AI Task's LLM receives descriptions of all connected tools and decides when and how to use them.
Email Tool
Icon: 📬
| Property | Required | Description |
|---|---|---|
mailConnectionId | Yes | Email integration ID |
recipientAddress | Yes | Recipient (AI fills this dynamically) |
subject | Yes | Email subject |
body | Yes | Email body |
contentType | No | text/plain or text/html |
The AI provides the specific values for recipient, subject, and body when it calls this tool.
SQL Tool
Icon: 🔍
| Property | Required | Description |
|---|---|---|
databaseConnectionId | Yes | Database integration ID |
mode | No | direct (AI writes SQL) or ai (natural language) |
question | No | Template question if using AI mode |
outputVariable | No | Variable name for results |
transactional | No | Wrap in transaction |
The AI generates the SQL query dynamically based on the user's question and the database schema.
REST API Tool
Icon: 🔌
| Property | Required | Description |
|---|---|---|
restApiConnectionId | Yes | REST API integration ID |
body | No | Request body template |
The AI determines the endpoint path, method, and body when calling this tool.
MCP Tool
Icon: 🔗
| Property | Required | Description |
|---|---|---|
mcpConnectionId | Yes | MCP server integration ID |
toolMethodsFilter | No | Comma-separated list of allowed MCP tool methods |
MCP (Model Context Protocol) allows connecting to any MCP-compatible server. The AI can invoke any of the server's registered tools.
Knowledge Base Tool
Icon: 📚
| Property | Required | Description |
|---|---|---|
collectionId | Yes | Vector store collection ID |
The AI searches the knowledge base with a semantic query generated from the user's question. Retrieved documents are returned to the AI as context.
This is the standard Retrieval-Augmented Generation (RAG) pattern:
- User asks a question
- AI generates a search query
- Knowledge Base Tool retrieves relevant documents
- AI generates an answer grounded in the retrieved content
Web Search Tool
Icon: 🔍
| Property | Required | Description |
|---|---|---|
provider | Yes | google or tavily |
apiKey | Yes | Search provider API key |
searchEngineId | Conditional | Google Custom Search Engine ID (required for Google) |
endpoint | No | Custom search endpoint |
The AI generates the search query and receives the search results as context.
File Manager Tool
Icon: 📁
| Property | Required | Description |
|---|---|---|
action | Yes | read, create, update, delete, or list |
fileId | No | File identifier (AI provides dynamically) |
filename | No | File name |
format | No | File format |
sourceFormat | No | Source format for conversions |
contentType | No | MIME type |
processFiles | No | Whether to process/parse file content |
Example: AI Support Agent with Multiple Tools
[AI Task: Support Agent]
│
├──toolConnection──► [Knowledge Base Tool: Support Docs Collection]
│
├──toolConnection──► [SQL Tool: Customer Database]
│
└──toolConnection──► [REST API Tool: Ticket System]
The AI receives the customer's question, searches the knowledge base for relevant articles, queries the database for the customer's history, creates a ticket via the REST API, and then generates a comprehensive response — all orchestrated automatically.
How the AI Uses Tools
The AI Task sends the LLM a description of each connected tool (name, description, parameters). When the LLM decides to use a tool:
- LLM outputs a tool call (JSON with tool name + arguments)
- Engine executes the corresponding Tool node
- Tool result is sent back to the LLM
- LLM continues its reasoning with the tool result
- Repeat until LLM generates a final text response
This loop is handled automatically by the engine — no manual wiring needed.
Best Practices
- Give each Tool node a descriptive name — the AI uses the node name as the tool description
- Connect only tools that are relevant to the AI's task — don't connect 10 tools if only 2 are needed
- For Knowledge Base tools, ensure the collection is up to date and relevant
- Use
toolMethodsFilteron MCP tools to restrict which server methods the AI can call - Monitor tool usage in execution logs — the AI Task logs each tool call and its result