A clustered filesystem can be very overwhelming to manage. Even routine operations like mounting filesystems, creating filesets, or creating snapshots require navigating chain of commands and dependencies. Trying to execute a simple operation needs understanding of the entire system topology, and relevant documentation.
What if this complexity could be reduced by just interacting with the cluster in plain english just the way we talk to each other? IBM Storage Scale solves this problem, by introducing Scale Provisioning Agent which provides a conversational interface for managing IBM Storage Scale filesets, snapshots, and other storage operations.
Built with LangChain and LangGraph, the agents integrate with the IBM Storage Scale MCP (Model Context Protocol) server to execute operations safely with user approval. In this blog, we’ll walk through the steps to setup the IBM Storage Scale agentic environment to get started with it.
What is a Provisioning Agent
The days of memorizing many commands with dozens of flags, or reading through long documentation and GUI pages belong to an older era. Everyone has now transitioned towards a conversational chatbots, ask it and get it. But unlike traditional chatbot-style interactions, this agent doesn’t just respond , it acts. It can call tools, run operations, and interact with the filesystems through a standardized interface, which is MCP.
IBM Storage Scale has its own MCP server with all the tools you need to manage the filesystem, and this provisioning agent is the missing link which simplifies the interaction with these tools. Since the MCP supports both transport modes, stdio and streamable HTTP, the agent and MCP server can reside together on a single machine, or could run independently.
And this does not mean the user lose control over his cluster. The user is very much in control with the human in loop design of the agent. Proceeding after the prompt requires a manual confirmation with a detailed response of what the agent is about to do. So the user knows what the agent is trying to do, making the agent actions more reliable.
Settting Up the Agent
Let us go through the steps to setup the Scale Provisioning Agent.
Prerequisites:
ollama pull qwen3:latest
Installation
-
Clone the repository:
git clone https://github.com/IBM/ibm-storage-scale-agents.git
cd ibm-storage-scale-agents
-
Install dependencies using uv:
uv sync
Or using pip:
pip install -e .
Configuration
Edit config/agents_settings.ini to configure the agent:
LLM Configuration
[llm]
model_name = ollama_chat/qwen3:latest
MCP Server Configuration
Option 1: HTTP Transport
[mcp]
transport = http
url = http://127.0.0.1:8000/mcp
Option 2: Stdio Transport
[mcp]
transport = stdio
command = /path/to/uv
args = --directory /path/to/scale-mcp-server run scale-mcp-server --transport stdio
Logging Configuration
[logging]
level = DEBUG
format = json
file_path = logs/agent.log
max_bytes = 10485760
backup_count = 5
Usage
Starting the Agent
Run the interactive CLI:
python main.py
Or using uv:
uv run python main.py
Once the setup is complete, the IBM Storage Scale Provisioning Agent is up and running.
Agent in Action:
Let us have a look at the Provisioning Agent in action, by creating a fileset.
Prompt: create an independent fileset myfset01 in the filesystem fs1
Figure 1: Prompt passed and manual confirmation for fileset creation
Agent Response:
Figure 2: Agent initiates the fileset creation
Prompt: list all filesets in filesystem fs1
Agent Response:
Figure 3: Agent response with list of all filesets with key note of the newly created fileset.
Prompt: list me the snapshots of fileset fset5 in the filesystem fs1
Agent Response: 
Figure 4: Agent response with list of snapshots for the required fileset
Conclusion:
With IBM Storage Scale Provisioning Agent powered by MCP, we offer a way to interact with tools that feels natural, flexible, and aligned with how we think and not how systems demand we think. This is not just a better user experience, it’s a fundamentally better way to work with complex systems. Feel free to use the agent and try out the agentic way of interacting with IBM Storage Scale.
#community-stories3