# AI Workflows The **AI Workflows** provides visibility into execution paths composed of agents, tools, and MCP servers. Each workflow is automatically discovered from connected environments, then visualized to show how components interact within the system.\ Alongside discovery, the feature detects **potential vulnerabilities** within workflows and provides **remediation guidance**, helping teams identify weak points and harden agentic architectures. {% hint style="info" %} AI Workflows is built on top of our successful open-source tool [**Agentic Radar**](https://github.com/splx-ai/agentic-radar), extending its capabilities into the enterprise. {% endhint %} ## Workflow Inventory

Figure 1: Workflow Inventory

The inventory table lists all discovered workflows with the following details: * **Name** – workflow identifier. * **Agentic Framework** – the framework used (e.g., OpenAI Agents, CrewAI). * **Scan Timestamp** – when the workflow was last scanned. * **Environment** – environment in which it was discovered. * **Source** – repository or project path where the workflow is located. * **No. of Agents** – number of agents included in the workflow. * **No. of Tools** – number of tools exposed in the workflow. * **No. of MCP Servers** – number of MCP servers integrated. From here you can search or filter workflows, or click on a workflow to explore its details. ## Workflow Details Clicking a workflow opens the **Workflow Details view**, organized into tabs: ### Overview

Figure 2: Overview Tab

* Displays workflow metadata (scan time, agentic framework, source). * Summarizes the number of agents, tools, MCP servers, and vulnerabilities detected. * Generates a **visual graph** showing all workflow nodes and their connections. * **Agents** – blue nodes. * **Tools** – purple and pink nodes (custom functionality). * **MCP Servers** – yellow nodes. The graph makes it easy to **visualize architecture, identify dependencies, and uncover risky connections**. ## Agents

Figure 2: Agents Tab

The **Agents tab** lists all agents within the workflow, with details including: * **LLM Model** – the model backing the agent. * **System Prompt** – the original system prompt defined for the agent. * **Hardened System Prompt** – the our hardened version with mitigations applied. Hardened System Prompt can be expanded for review and copied directly from the interface. ## Tools

Figure 3: Tools Tab

The **Tools tab** shows each tool available to agents in the workflow: * **Tool Name** and **Category** – e.g., WebSearchTool under web\_search. * **Tool Description** – functionality and supported operations. Clicking on **Show Vulnerabilities** opens a vulnerability details modal:

Figure 4: Tool Vulnerabilities

* **Description** – how the issue can be exploited. * **Security Framework Mapping** – mapped security framework items. * **Remediation Steps** – actionable recommendations to reduce risk. ## MCP Servers

Figure 5: MCP Servers Tab

The **MCP Servers tab** lists all Model Context Protocol servers discovered in the workflow, with configuration.\ This highlights external integrations and the commands or services exposed through MCP.