Agents & MCP Tools

​

Overview

Agents help make workflows autonomous. An Agent can decide what tools it needs to perform a task, combine those tools, and consolidate the results into a final outcome. This is powered by MCP (Model Context Protocol) servers, which expose external systems - like SaaS apps, APIs, or internal services - as callable tools. Agents don’t talk to these systems directly. Instead, they interact through a clean, governed tools layer. This setup gives you:

• Workflows that can adapt at runtime
• More controlled access to external systems

​

Core Building Blocks

The Agents & MCP system is built around three main components:

• MCP Servers - Helps AI agents perform actions through pre-defined tools
• MCP Tools Component - A way to select and grant tool access to an Agent
• Agent Component -Selects tools and executes actions

​

MCP Servers

An MCP Server lets an Agent take actions in external systems, like searching data or creating records. AI Squared supports two ways to work with MCP servers:

• Preconfigured servers
• Custom servers 

Each MCP server has connection details and authentication settings.  Before a server can be used, the connection is tested to ensure it is valid and reachable, and exposes valid tools.

​

MCP Tools Component

The Tools Component defines which MCP tools an Agent can use in a workflow. The MCP tools themselves do not execute by themselves. The Tools Component:

• Points to an MCP server
• Lists the tools available by that server
• Lets users select which tools an Agent is allowed to use

This keeps closer access control over the tools.  What’s Available:

• 26 pre-integrated remote MCP server tools
• Custom MCP Server support to connect any MCP-compliant server
• Centralized tool configuration and management
• Tools configured today can be reused in upcoming workflow and agent capabilities

​

Agent Component

The Agent Component is where execution actually happens. An Agent:

• Uses an LLM to reason
• Decides which tool to use based on the task
• Calls tools through the MCP client
• Interprets tool responses
• Produces a final result

Agents run on the backend and support multi-step reasoning, making them ideal for workflows in the real-world workflows.

​

How the Workflow Functions 

When a workflow runs:

1. The scheduler reaches the Agent
2. The Agent inspects which tools are connected
3. The Agent builds an internal list of allowed tools
4. The Agent decides which tools to call and when
5. Tools are invoked dynamically through the MCP client

The workflow engine does not execute tools directly - only the Agent does.

​

Safety, Control, and Governance

Several guardrails are built into the system:

• Agents can only use tools connected to them
• Tool credentials are stored securely
• Logs are captured for debugging

These controls ensure autonomy without compromising on security.

​

Why Agents & MCP Tools are Important

Agents and MCP Tools together help make workflows smarter, compliant and efficient. They allow teams to:

• Automate complex, multi-step tasks
• Integrate AI reasoning with real systems
• Keep tool access explicit and auditable
• Scale your operations efficiently.