How to build Node.js MCP Servers that Expose Data Resources to LLMs

MCPs are famous for exposing Tools that augment Large Language Models (LLMs) with the ability to perform actions and invoke generic functions that MCP Servers can implement, but exposing data resources to LLMs is just as important.

What are MCP Server resources?

Resources allow MCP Servers to expose data as read-only information that LLMs can use as context and query for data points. For example, an MCP Server can define database records as resources, an application’s status, or any other form of text or binary data content.

Practical examples of MCP Server resources include:

• An organizational org chart for a company.

• A documentation server exposing API data.

• Product FAQ and troubleshooting guidelines.

MCP Server resources are defined using a unique URI and will often contain their own custom protocol scheme. Building on the examples above, an MCP Server resource for an organizational org-chart could be org-chart://company-wide, org-chart://engineering, org-chart://cto and so on.

These URI requests are sent from the MCP Client to the MCP Server, which needs to define specific routes for these as part of the server.setRequestHandler MCP Server API.

Building a Node.js MCP Server for resources

Let’s build an MCP Server in JavaScript to run with a Node.js runtime and define Resources as part of this MCP Server data points.

Begin by creating the file ./resources/index.js, which hosts the MCP Server Resources definition and to which we will begin by importing the necessary code from the Model Context Protocol and a handy logger utility:

Next, we will define our main exported function that creates the resource. In this function, our Resource is going to be an open source SVG representation of the Node.js runtime release schedule. We can then query for questions like when is the next LTS version of the Node.js runtime planned for?

Let’s implement the logic to fetch this data:

The above code simply fetches the data, includes generic error handling, and creates the resource as text data from the SVG file.

Next, we define an array of resources. Each MCP Server Resource definition has the following properties:

• A URI.

• A name.

• A description.

• A mime type.

We also defined a function handler so we can dynamically resolve all the resources. Following is the continuation of our initializeResources function, which uses the server.setRequestHandler API to list the resources as well as implement them by terminating the route on the request.params.uri from the incoming request (originating from the MCP Client) and then invoking the function handler in our resource:

Now let’s put this together via two files: index.js and server.js, which will expose the MCP Server, and then wrap all of this up as a generic Node.js application process:

The server.js, as described above, will declare the MCP Server as STDIO transport and expose it:

Then the index.js file imports the MCP Server definition and starts:

Building MCP Servers

If you want to gain a better understanding of how MCP architecture works, its main components, and the differences between MCP to REST and MCP to Function Calls, then I highly recommend you follow up with the beginner's guide to visually understanding MCP architecture.

If you’re entirely new to MCPs, check our article What is MCP in AI? Everything you wanted to ask. It is a great start to gain fundamental knowledge of what MCPs solve.

My last recommendation is that if you choose AI coding tools to write code, like Cursor and others, I invite you to add the Snyk CLI MCP Server, which will help provide context to your agentic code workflow about insecure coding patterns and vulnerable dependencies that can mitigate security issues in an automated way.