A Sneak Peek at the Pure Fusion MCP Server

Pure Fusion™ already gives you a powerful, fleet‑wide control plane for FlashArray™ and FlashBlade®. Now, there’s a new way to plug that context directly into your AI workflows: the Pure Fusion Model Context Protocol (MCP) server.

In just a couple of weeks, we’ll make the Pure Fusion MCP server generally available as an open‑source project in the Pure Storage OpenConnect GitHub organization, so customers and partners can inspect the code, extend it, and integrate it into their own automation frameworks.

This post is your “sneak peek” of the Pure Fusion MCP Server. Stay tuned as we go deeper into the architecture, setup, and real‑world workflows.

What is the Pure Fusion MCP Server?

At a high level, the Fusion MCP server is a lightweight service that lets any MCP‑aware AI assistant talk to your Pure Fusion environment using a small set of well‑designed tools instead of low‑level REST calls.

How we built our own MCP server

The server is designed with a few key principles in mind:

• Thin, stateless front end over existing APIs
  It doesn’t reinvent storage control. Instead, it layers on top of the FlashArray and FlashBlade REST APIs and Pure Fusion’s control plane, translating natural‑language intent into a handful of higher‑level tools like “query fleet status”, “list workloads”, and “get performance metrics.”
• Downloadable, self‑contained binary
  You run it as a small cross‑platform binary (macOS, Linux, Windows) on an admin workstation, a jump host, or a small VM/container.
• Simple, token‑based auth to your arrays
  The server uses API tokens you generate on your FlashArray and FlashBlade systems. A built‑in command helps you validate tokens, detect API versions, and produce a usable config in one go.
• Read‑only, insight‑first workflows
  The initial focus is on passive, read‑only use cases: fleet and array status, configuration discovery, inventory, and telemetry queries. That makes it safe and easy to introduce into existing automations and processes.
• Open protocol, open source
  It implements the Model Context Protocol (MCP), the emerging standard for connecting AI agents to external systems. Pure Storage is releasing the Pure Fusion MCP server as open source in the PureStorage‑OpenConnect GitHub organization so you can fork, inspect, and contribute just as you do with our other OpenConnect projects.

How will customers and partners use Pure Fusion MCP?

Because it speaks MCP, the PureFusion MCP server can plug into desktop AI assistants, CLI agents, IDE plugins that understand MCP, and any custom internal agents that your automation or SRE teams build

You can query and ask fleet‑level questions about health, capacity, and performance instead of scripting every call. Or, quickly spot drift in NTP, DNS, LDAP/IdP, SafeMode™, and other settings across dozens of arrays. It will also provide the ability to map Pure Fusion workloads, presets, and volumes to applications and hosts without endless cross‑referencing. Finally, combine Pure Fusion MCP with other workload MCP servers to orchestrate realistic end‑to‑end demos from a single prompt. Let’s take SQL as an example.

I know what you’re saying. “I get it, but what could it do for me, my team, my users, and especially, my boss?” How about we walk through a scenario that shows why both storage admins and DBAs will care about MCP—and how you can combine the Pure Fusion MCP server with a Microsoft SQL Server MCP server to get a complete picture.

Let’s talk about the problem

You have a mission‑critical SQL Server workload that has multiple FlashArray systems in a Pure Fusion fleet hosting SQL data and log volumes, a SQL Server running on Windows or Linux VMs, with several large databases. Users are complaining about intermittent slow queries during peak hours.

Traditionally, troubleshooting this means:

• Storage admin digging into array performance graphs, volume stats, and host configuration
• DBA looking at waits, missing indexes, query plans, and file layouts inside SQL Server
• A lot of context‑sharing over tickets and screenshots, and manual correlation of “this volume on that array” with “this file in that database.” 

It’s all the right work—just spread across tools and teams. And that takes time, and resources. Valuable things not all admins have, am I right?

What’s the solution?

With MCP, both sides can lean on a shared AI agent that has two servers connected. In this example, The Pure Fusion MCP server for storage and fleet context., and the Microsoft SQL Server MCP server that understands databases, files, waits, and query performance.

Here’s what an example workflow might look like:

• The storage admin kicks things off. They open their MCP‑aware AI agent like VSCode or Claude and ask:

“Using the Pure Fusion MCP tools, show me all Pure Fusion workloads and volumes that host our ProdSQL environment, including their arrays, IOPS, latency, and bandwidth over the last two hours.”

The agent then uses the Pure Fusion MCP server to enumerate workloads, presets, and volumes related to ProdSQL. It pulls per‑volume and per‑array performance metrics from the relevant arrays, and summarizes which volumes or arrays show latency spikes during the reported slow periods.

• Now, it’s the DBA’s turn. The DBA asks the same assistant: “Using the SQL Server MCP tools, correlate those storage volumes with SQL Server database files, list the top databases by I/O during those windows, and summarize the main wait types.”

The agent again uses the SQL Server MCP server to map storage volumes to database files (data, log, tempdb). It then identifies which databases were hottest and which wait types (e.g., I/O‑related waits) dominated during the slow periods.

• The agent now brings it all home. With both views available, the assistant can now answer questions like:

“Which SQL databases on ProdSQL are most affected when array FA‑X01 shows high latency?”

“Are the busiest databases sharing the same underlying volumes or arrays?”

“Is there evidence of tempdb or log contention that aligns with storage hot spots?”

So, what did this solve? Instead of manually correlating volume and host mappings, Array performance charts, and SQL waits and file layouts, Both teams see a single, AI‑generated narrative that points to likely causes—whether it’s a particular volume under pressure, a misaligned file placement, or a noisy neighbor workload.

The key idea: the Fusion MCP server gives your AI agent deep fleet and storage context, while the SQL MCP server provides application‑level insight. You still decide what changes to make—but the time to diagnosis drops dramatically.

What’s next for the Pure Fusion MCP server and MCP-Based Automation

This is just the beginning. This post is the first in a technical series on the Pure Fusion MCP server. In upcoming posts, we’ll get into Installation and configuration with step‑by‑step guidance and troubleshooting tips for real environments. How about some deep dives? Yep, we’ll show how individual Pure Fusion MCP tools map to Pure Fusion and array APIs, and how to combine them safely. Finally, let’s conjure up some multi‑MCP workflows using Pure Fusion MCP alongside other MCP servers to build rich automation and solutions.

So, get up to speed on MCP if you don’t know about it already by reading the Model Context Protocol overview to understand how MCP servers, tools, and clients fit together:
https://modelcontextprotocol.io/docs/getting-started/intro

Stay tuned for the next post in this series will take you from download to your first successful MCP query against your Pure Fusion fleet, with clear, copy‑and‑paste examples you can try in your own environment.