Bridging the gap between AI tools and enterprise data

The AI revolution has reached a critical inflection point where enterprises need seamless access to their proprietary data to unlock AI’s full potential. However, enterprises struggle to integrate proprietary data with generative AI (GenAI) applications in a way that’s simple, secure, and cost-effective. By extending our innovations over the past year to help transform enterprise data into searchable knowledge, today we’re pleased to announce the release of the NetApp^® Model Context Protocol (MCP) Server for Knowledge Bases. This MCP server simplifies integration of the insights that are derived from enterprise data with AI applications.

Traditionally, the integration of GenAI applications with up-to-date and context-rich enterprise data requires complex, custom connectors. With a growing number of applications and the data sources that those applications need to access, managing such integrations can be highly complex. MCP reduces this complexity by providing a standardized protocol that enables GenAI applications to dynamically access data sources, such as databases, file shares, and API resources, without custom integrations. The following figure shows an example of MCP architecture. This mechanism offers a standard, simple, and scalable way for applications to provide context to large language models (LLMs).

The preceding figure shows how an application such as Claude Desktop that’s running locally on a user’s machine can use MCP to communicate with an MCP server. For example, MCP Servers A and B run on a user’s local machine with local data sources. MCP Server C runs locally on a user’s machine that’s connected to a remote data source location. Also shown is remote MCP Server D, which is connected to a remote data source. 

This approach exposes certain capabilities so that the application can add context to the models that the application is using. For example, Anthropic^® Claude Desktop can query a local database or a remote service such as NetApp BlueXP™ workload factory to search for context that’s relevant to a user’s query and then add that context to answer user queries.

Before we go further, as described in the MCP specification, we should understand a few key components:

• MCP hosts and clients. Hosts are applications such as Claude Desktop, Amazon Q, Cursor, and any other AI tools that want to access data by using MCP. Clients are connectors within the host system that maintain connections with servers.
• MCP servers. Servers are lightweight programs that each expose specific capabilities through the standardized MCP. You can deploy MCP servers locally on the host machine or access them remotely.
• Local data sources. Local sources are your computer’s files, databases, and services that MCP servers can securely access.
• Remote services. These services are external systems that are available over the internet (for example, through APIs) that MCP servers can connect to.

Two other important concepts also define how models retrieve the context and how they behave:

• Tools enable servers to expose executable functionality to clients. Through tools, LLMs can interact with external systems, perform computations, and take action in the real world. Tools are designed to be controlled by the models so that the models can automatically invoke them. The clients can discover the tools that are exposed by the MCP server and invoke them as necessary. For example, tools can include querying databases, calling APIs, writing files, or triggering workflows.
• Resources are a core primitive in the MCP that enable servers to expose data and content that can be read by clients and that can be used as context for LLM interactions. Unlike tools, resources are designed to be controlled and invoked by the application instead of by the models. Resources can be documents, databases, or any data source that the AI needs to access for information. Also, resources can be used according to the logic that’s built into the application instead of being used by the models at run time.

Now that we understand the basics of MCP, let’s look at how the NetApp MCP server helps bridge the gap to connect GenAI applications to knowledge that’s derived from enterprise data sources.

Today, organizations are seeking ways to transform their collective knowledge into actionable insights that are accessible directly within their applications. A lot of an organization’s knowledge is captured in enterprise datasets—especially unstructured data on enterprise storage systems.

With a vision to help customers like you access this vast knowledge in a simple, secure, and cost-effective manner, NetApp launched the BlueXP workload factory for GenAI service. You get a managed retrieval-augmented generation (RAG) service to connect your corporate file systems (resident on NetApp or any SMB/NFS storage) and to transform those datasets into structured, queryable knowledge bases by using natural language. You can integrate these knowledge bases with GenAI applications by using the REST API to search (with the knowledge base search API). You can then retrieve relevant context for a user query or generate a response to the user query (by using the chat API). The following figure shows an example use of NetApp knowledge bases.

Although your developers can continue to use our REST API for integration, we have made this process even simpler—with a local MCP server. The local server exposes knowledge bases as MCP tools that end-user applications can access to retrieve relevant context for user queries. In our first implementation, you can install a local MCP server on a host machine that’s running applications such as Claude Desktop and that’s connected to NetApp knowledge bases (like MCP Server C in the first figure). You can download and install the MCP server on a host machine and configure it to allow the user access credentials for BlueXP. For Claude Desktop users, we have also created the desktop extension package for single-click installation and for securing secrets through Claude Desktop’s use of a local host OS keychain. 

After it has been installed, the MCP server exposes all configured knowledge bases as MCP tools, along with a description of each knowledge base. The models that the application uses, such as Anthropic Claude Sonnet, can then determine which knowledge bases need to be invoked to search for context for user queries. The model uses knowledge base descriptions as additional information to understand how to use each knowledge base.

The following figure illustrates how user queries are routed through the MCP server to the BlueXP workload factory knowledge base. Based on the query, the model selects one of the tools or the knowledge bases to search for the relevant context to answer the user’s query. The model then invokes the knowledge base search API to search for and to retrieve the top relevant chunks from the indexed data sources. Up to the top 10 chunks, derived through a hybrid search (vector and full text) and ranked by a reranking model, are returned as part of the search results and are provided to the model. The model then generates the answer to the user query by using the retrieved chunks as context.

Use case 1: Claude Desktop for a business knowledge worker

Let’s consider a utilities company where an energy pricing professional needs rapid access to oil and natural gas production and forecast data to advise procurement decisions. With the NetApp MCP server, Claude Desktop can seamlessly query the company’s knowledge base, which contains the company’s internal and market reports and forecast data. When the professional asks Claude about price forecasts, the AI assistant can instantly retrieve relevant information from the documents that are indexed in the knowledge base to quickly answer the user’s question. This capability can vastly improve productivity and efficiency of knowledge workers by instantly giving them the needed information to complete their tasks.

To illustrate the above example, view this video for a demonstration of how Claude Desktop can integrate with knowledge bases using the MCP server.

Use case 2: Amazon Q Developer for accelerated software development

Development teams who are working on complex enterprise applications often need access to internal documentation, API specifications, coding standards, and so on. The NetApp MCP server for knowledge bases enables developer assistants such as Amazon Q Developer to access the organization’s technical knowledge base which could include design and architectural documents, API specifications, and troubleshooting guides, etc. When developers need help in developing code, when they encounter issues, or when they need to understand existing systems, these tools can automatically retrieve relevant information and even generate code using the derived knowledge.

This integration dramatically reduces the time that developers spend on searching for information, and it helps maintain consistency across development teams. New team members can use the collective knowledge of the organization immediately, reducing onboarding time while maintaining adherence to established coding standards and architectural patterns.

With the launch of the NetApp Knowledge Base MCP Server, we are moving toward making enterprise AI more accessible and practical. By standardizing how AI tools access enterprise knowledge, your organization can avoid the fragmentation and technical debt that often accompany custom integrations.

Early adopters are already seeing significant returns on their investment, with organizations reporting improved decision-making speed, reduced training costs, and enhanced innovation capabilities. The ability to provide AI tools with contextual, up-to-date information from enterprise knowledge bases creates a competitive advantage that compounds over time.

As enterprises continue to embrace AI technologies, the ability to seamlessly integrate proprietary knowledge with AI tools will become increasingly critical. The NetApp Knowledge Base MCP Server is just the starting point to simplify access to enterprise knowledge.

Recently, the MCP server specification has added support for enterprise-class external authentication and authorization through OAuth. Knowledge bases already support OAuth integration, along with granular permissions control, so that only authorized users can access the knowledge bases and so that responses adhere to the original document permissions. In future releases, we’ll add support for external authentication and an option to run the MCP server remotely. Applications and authenticated users can then securely connect to a centralized MCP server and gain access to all the knowledge base tools that are available to add context to LLMs.

The future of enterprise AI lies not in replacing human expertise but in augmenting it with intelligent access to organizational knowledge. The NetApp MCP server for Knowledge Bases makes this vision a reality, enabling your organization to transform your data sources from static repositories into dynamic, AI-accessible resources that drive innovation and a competitive advantage.

To get started, review more details in our documentation and download the MCP server from our GitHub repository.

Authors: Puneet Dhawan, Yuval Kalderon, Alexandar Korman, Chen Bassat, Michael Shaul