In an earlier blog post describing experimentation with AI capabilities in HPCBOX here, we looked at the possibility of using an MCP web service. Over time, we’ve made a pivot and concluded that it is more productive to empower users with AI capabilities directly within the HPCBOX cluster desktop and a local stdio MCP server per user.
Design
We’ve decided to add AI capabilities directly into the HPCBOX platform and tightly integrate the HPCBOX Workflow Designer with the AI assistant. We are also exposing an MCP server for external agents while keeping the core implementation shared between the agent and the MCP server. Additionally, we are currently using Microsoft Agent Framework which lets our users select the LLM of their choice, including self-hosted LLMs with Ollama.
In this new design for AI capabilities in HPCBOX, we’ve also added context memory for assistants to remember what a user worked on in the past and also have a shared memory context setting which lets groups of users share their context memory. This shared memory context is very beneficial for a group of users working together on CAE, ML or AI projects and lets them share best practices, status of projects etc. This also allows users to share commercial license usage plans with each other on an HPCBOX cluster.
Built-in AI Assistant in HPCBOX
The HPCBOX AI Assistant is an embedded chat panel in the workflow editor that helps users create and modify workflows using natural language. Instead of manually dragging steps, configuring options, and drawing connections, they can describe what is to be built and the assistant builds it.
MCP server for external agents
HPCBOX includes an MCP server, which is the same binary as the built-in agent, that lets external AI agents create and manage workflows on the cluster. This means users can use their preferred AI tool — Claude Code, VS Code Copilot, Cursor, or any MCP-compatible client — to build HPCBOX workflows.
The HPCBOX MCP server exposes workflow management tools via stdio transport. AI agents connect to it, discover the available tools, and call them to list applications, create workflows, validate structures, and more.
Interactions
Some example interactions of how AI can be used in HPCBOX to build HPC workflows for CAE, AI or ML applications or even workflows that are a combination of them.
Using the AI Assistant to create an HPC workflow
In the example below you can see me creating an ANSYS workflow with CFX and CFD-Post
Shared memory recall in the AI Assistant
In the example below, you can see me asking the AI Assistant to update me with projects that other users have created on the HPCBOX cluster.
Using the MCP server from an external agent
In the example below, you can see me using VS Code and Copilot to interact with the HPCBOX Workflow Designer and create a workflow by pointing to an OpenFOAM tutorial link. This capability is cool since the MCP server could be used even by other agents like those which might have been created on Microsoft Foundry in Azure using ssh+stdio support.
Conclusion
The new AI capabilities in HPCBOX empower users with increased productivity in this age of automation, expanding HPCBOX platform capabilities and, making it usable from other AI systems which would need to make use of HPC in the larger context of the problems they are solving.
HPCBOX AI Assistant with Shared Context Memory - YouTube
HPCBOX AI Assistant with Shared Context Memory
AI Agent functionality in HPCBOX - YouTube