As AI becomes a first-class participant in enterprise operations, we’re starting to rethink how we interact with complex infrastructure. Instead of writing scripts and making endless API calls, what if we could simply ask our systems questions in natural language and get reliable, grounded answers? 

One of the most promising steps in this direction is the Model Context Protocol (MCP). In this blog, we’ll walk through how to set up an MCP server for IBM Storage Scale, and connect it to two different types of clients: 

Before diving into commands and configuration, let’s discuss why this matters, especially for those of us managing IBM Storage Scale environments. 

Why This Matters 

Managing a large IBM Storage Scale deployment often means checking cluster health, inspecting node status, and diagnosing issues via REST APIs. While powerful, this process is manual, you must know the endpoints, execute calls, parse responses, and connect the dots. Imagine instead asking, “What is the health of my cluster?” and getting a clear, trustworthy answer instantly. 

That’s exactly the gap MCP is designed to close. 

Where MCP Fits In 

MCP leverages the APIs and exposes them to AI systems in a structured and controlled way. Instead of free-form responses, MCP defines explicit operations called tools that an AI assistant is allowed to execute. 

These tools map directly to IBM Storage Scale capabilities, such as: 

In this article, we focus primarily on tools, as they are the key mechanism that allows AI assistants to interact meaningfully, and safely, with your Storage Scale environment. 

Installation Guide 

Prerequisites 

Before you begin, ensure you have the following: 

curl -LsSf https://astral.sh/uv/install.sh | sh

Setting Up the Scale MCP Server

1. Clone the Scale MCP Server Repository 

Start by cloning the official Scale MCP repository: 

git clone https://github.com/IBM/IBM-Storage-Scale-MCP-Server 

cd scale-mcp-server 

2. Set Up Python Environment and Install Dependencies 

Use UV to create a virtual environment and install the required packages: 

uv venv 

source .venv/bin/activate 

uv pip install . 

3. Configure Scale Connection Settings 

Edit the configuration file at config/scale_config.ini with your IBM Storage Scale cluster details: 

source .venv/bin/activate 

scale-mcp-server --transport stdio 

    Figure 2: Scale MCP server running using stdio 

Connect Claude Desktop as MCP Client (stdio) 

This section walks through connecting Claude Desktop to the Scale MCP Server using the stdio transport for local, process based communication. 

To connect to Claude desktop, follow the below steps: 

    Figure 3: Configuring Claude Desktop to connect to the Scale MCP Server using stdio transport

{ 

  "mcpServers": { 

    "scale-mcp-server": { 

      "command": "/path/to/scale-mcp-server/.venv/bin/scale-mcp-server", 

      "args": [ 

        "--transport", 

        "stdio" 

      ] 

    } 

  } 

} 

          In Claude Desktop, go to Settings → Developer, and you should see your Scale MCP server listed as running. 

    Figure 4: Claude Desktop successfully connected to the Scale MCP Server

          Once connected, you can ask Claude about your Storage Scale environment. For example: 

          “Can you give information about my cluster?” 

    Figure 5: Asking Claude a natural language query about the IBM Storage Scale cluster

    Figure 6: Asking Claude a natural language query about the IBM Storage Scale cluster 

The MCP server will handle the request internally using IBM Storage Scale tools, giving you instant, reliable insights. 

Connecting IBM Bob as an MCP Client (StreamableHTTP) 

In addition to Claude Desktop, the Scale MCP Server can also be consumed by IBM Bob or any MCP compatible web client using the StreamableHTTP transport. This is useful for browser based or hosted AI interfaces where stdio is not applicable. 

Prerequisite 

Ensure that your Scale MCP Server is running with HTTP transport enabled: 

uv run scale-mcp-server 

By default, the MCP server will be available at: 

http://127.0.0.1:8000/mcp 

    Figure 7: IBM Bob home screen showing the option to connect to an MCP server

    Figure 8: MCP server configuration

         Update the mcp_settings.json file as shown below, using streamable-http and pointing to the local MCP server URL: 

{ 

"mcpServers": { 

"scale-mcp-server": { 

"type": "streamable-http", 

"url": "http://127.0.0.1:8000/mcp" 

} 

} 

} 

Save the file once the configuration is updated. 

    Figure 9: Shows the MCP server listed

          Once connected, you can begin asking Bob questions about your IBM Storage Scale environment, such as: 

          “Can you list all linked filesets” 

    Figure 10: Asking Bob a natural language query about the IBM Storage Scale cluster 

Figure 11: Asking Bob a natural language query about the IBM Storage Scale cluster  

This completes the setup, enabling IBM Bob to query and interact with IBM Storage Scale via the MCP server. 

Conclusion 

Connecting Claude, Bob or any MCP compatible client to your IBM Storage Scale environment brings conversational AI directly into enterprise storage management. By exposing structured tools via MCP, you can monitor clusters, check health, and query metadata without manually navigating APIs. 

If you work in Storage, AI, or DevOps, I encourage you to: 

Experiment with extending these tools to streamline workflows, improve efficiency, and unlock new possibilities in storage management.