Understanding AI Agent and Tool Execution

The following section explains required configurations of AI Agent step, such as configured AutomationEdge workflow as tool, understanding of MCP server tools, also explains the step execution flow with an example.

AI Agent: An AI agent autonomously analyses prompts and uses tools for context-driven Agentic AI automation, enabling it to pursue goals, make decisions, and adapt intelligently. By using AutomationEdge workflows as tools or MCP tools, AI agents can perform tasks autonomously and accomplish their objectives.

What is MCP Server

An MCP server allows us to connect with external tools, workflows, and services using a standardized protocol. It acts as a bridge between the client (such as an AI system) and external tools or services, enabling reliable and consistent tool execution. To use an MCP server, configure the server connection in the Agentic AI setup. Once configured, the Agentic AI step selects the appropriate MCP server based on the user prompt and the tool description. See Tool Configuration tab section. When the Agentic AI step receives a prompt, it interprets the request and identifies the required capability. The step then invokes the MCP server, which executes the associated tool or workflow and processes the request. After execution, the MCP server returns a structured response to the Agentic AI step.

Memory and Knowledge Base for Tool Execution

The agent can use AE tools or MCP tools to perform tasks autonomously, guided by memory and knowledge.

• Knowledge Base: The knowledge base provides contextual information that AI agents use to perform tasks and call the appropriate tools with necessary parameters.
• Memory: The AI agent uses a conversation ID to store user messages and tool responses in memory. This enables the agent to maintain context across the conversation and use past interactions to select the appropriate tools and provide accurate parameters when executing them.

How to convert AE Workflow into Tool

• To use workflow as tool in AI Agent step, the Agentic AI Tool configuration must be configured in AutomationEdge server:
  Agentic AI Tool configuration:

  1. Configure the following properties in the ae.properties file as required:
     ae.feature.agentic-ai-tools.enabled=true
     ae.synchronous.wf.execution.api.timeout.seconds=60
  2. Ensure the agent thread count is set to more than 2 to allow proper execution of the AI agent and its tools on AE.

• Always define the tool name and description clearly. This helps the AI agent understand the tool’s purpose and choose it correctly when completing a specific task.

• Set workflow runtime parameters, which will be treated as AE tool input parameters.

• Set the workflow result using the Set Workflow Result step so the agent can receive the tool result and take appropriate actions based on it.

How AI Agent uses tools

When executing a tool, the AI agent determines parameter values using multiple sources mentioned as follow: User Prompt: Extracts explicit values directly mentioned by the user in user prompt.

Knowledge Base: If parameters are not fully specified in the prompt, the agent retrieves relevant information from the configured knowledge base.

Conversation Memory (If enabled): Use values from past interactions within the same conversation ID.

See the following example to understand execution flow of the AI Agent step:

This AI agent acts as an IT Assistant that helps diagnose and resolve technical issues, answer IT-related queries, and provide step-by-step solutions.

System Prompt: You are an IT Assistant. Your job is to help users diagnose and resolve technical issues or fulfil IT related requests. You can answer questions, provide troubleshooting steps, escalate problems when needed, and give clear, concise responses. Stay professional, friendly, and solution focused. Ask clarifying questions if needed. Thought: You should always think about what to do and list down actions before executing them. Action: The action to take, execute tools as necessary relevant input parameters

User Prompt: OpenKM Application is DOWN. AE Tools: Ping: This tool executes a ping command to check network connectivity with the specified host. Traceroute: This tool executes a traceroute command to identify the network path and detect potential connectivity issues. Note: Input parameter for both tool is host ip.

Knowledge Base:

Steps (Tools)

1. Ping the Impacted CI: Run ping 10.41.5.6 to verify network connectivity to the OpenKM server.
2. Traceroute to the Impacted CI: Use traceroute 10.41.5.6 to identify network bottlenecks issue.

AI Agent Tool Execution Logic: This AI agent has two diagnostic tools available: Ping and Traceroute. When an incident is reported (for example, “OpenKM Application is DOWN”), the agent follows the predefined steps from the Knowledge Base:

1. Ping Tool a. The agent first calls the Ping tool using the impacted CI’s IP address. b. This checks basic network connectivity to confirm whether the server is reachable.
2. Traceroute Tool a. If required, the agent then calls the Traceroute tool with the same IP address. b. This identifies the network path and helps detect routing issues or network bottlenecks.

The agent executes the tools in sequence, strictly following the steps defined in the Knowledge Base, and uses the tool outputs to determine the next action or escalation.

• AI Agent Response: I have performed a ping and traceroute to the OpenKM server.
• Ping to 10.41.5.6 was successful.
• Traceroute to 10.41.5.6 shows poor connectivity, which could indicate a network issue.