Improved Agentic Infrastructure

When users interact with AI in the TotalAgility chat interface, MCP tools may need to be executed to complete tasks such as reading files or accessing external systems. Without consent controls, these tools could execute automatically, creating security and compliance risks.

The MCP user consent feature adds a permission layer that:

• Gives users granular control over tool execution (deny, approve once, approve the session, or approve always).
• Allows users to manage their tool permissions through their profile settings in the Workspace.
• Enables administrators to monitor and revoke approval for a server, tool, and user.
• Logs all consent decisions and tool executions for audit and compliance.

The Generative AI activity is extended to include MCP support.

TotalAgility provides a default MCP Server within integrations where users can expose processes and agents as tools. The LLM invokes them based on the process name and description.

This capability allows TotalAgility processes and agents to be invoked directly using natural language, without requiring users to know the underlying process names, parameters, or how to trigger them manually. By exposing a process as an MCP tool, the LLM can understand what the process does (from its name and description) and automatically determine when and how to call it based on the context of a conversation.

The MCP tool selection interface has been improved for usability when working with MCP servers that expose large numbers of tools. When overriding MCP tools at design time in Generative AI chat control, AI Knowledge Base, and Copilot, you can:

• Use the Search tools box to find tools by name.
• Use the All/Allowed list to display tools from a specific server.
• Combine search and server filters to narrow tool selection.

Refer to Configure MCP server connection.