Improved Agentic infrastructure

The Agentic infrastructure in TotalAgility 2026.3 offers enhanced MCP, multiple models in AI Proxy models, and system security prompts in LLM calls.

Enhanced MCP support

TotalAgility's MCP support provides greater control over the tools available on third-party MCP servers. This includes managing tool availability and ensuring tool usage is fully auditable, enabling better governance and oversight.

• The new Allow all tools property, available in MCP server integration, allows you to control which tools are available for use. While all tools are enabled by default, you can deactivate any tools you don't need or don't want users to access. At runtime, only the tools you choose to enable will be available to users.

• At design time, when Use MCP servers is enabled within chat for Generative AI, AI Knowledge Base, and Copilot, all tools exposed by the selected MCP servers are available by default.

  • You can dynamically override MCP tools in a chat. Using the Override tools option, you can specify which tools are available for use within the chat at runtime.

  • Show MCP tool progress within chat. When enabled, the new Show tool progress property displays the tool name when generating a response, indicating that MCP tool processing is in progress during chat interactions.

• Audit the MCP tools used within chat interactions. When the Audit conversation option is enabled, each tool usage during a chat interaction is recorded in the audit log, enabling complete traceability of tool execution.

Support for multiple models

The Tungsten AI is extended to enable selection of your preferred supported model for each of the following features:

• Copilot for development text

• Copilot for development vision

• Copilot for extraction

• Knowledge Base search

Models listed are those made available for use by Tungsten Automation.

If a model selected by the customer in System settings is no longer supported within the proxy, the default model is used.

System security prompt

A new System security prompt setting under system settings enables you to customize a security prompt that is used when any call to AI is performed.

This setting helps enforce organization-wide security guidelines at the system level, which the users cannot modify or override, ensuring that all prompts are processed in accordance with organizational policies, regardless of inputs provided by end users. The setting does not apply to "Capture for extraction."