MCP Server - A Node In Layers Package for building MCP Servers
This library adds the ability to easily create MCP servers with Node In Layers.
It has a companion library called '@node-in-layers/mcp-client' which is used for creating MCP clients. These two libraries share the same functions for defining models and tools.
New Layer
This library adds a new layer mcp to the system. It should be placed after the express layer.
Usage
In order to use this library, you must make additions to your config, as well as create and export "mcp" layers from your apps/domains.
Config
you add this app/domain to your config file. You should do this before your apps which will add tools to the MCP server.
You then configure the mcp app/domain with the following:
const mcpConfig = {
// (optional) The name of your MCP server.
name: 'mcp',
// (optional) The version of your MCP server.
version: '1.0.0',
// The server config from @l4t/mcp-ai/simple-server/types.js
server: {
connection: {
type: 'http',
host: 'localhost',
port: 3000,
},
},
logging: {
// optional
// If you want to change the default. Its 'info' by default.
requestLogLevel: 'info',
// If you want to change the default. Its 'info' by default.
responseLogLevel: 'info',
},
}
const config = {
['@node-in-layers/mcp-server']: mcpConfig,
}
Creating an MCP Layer
You can create an MCP layer by exporting a function from your app/domain that returns a layer.
// /src/yourDomain/mcp.ts
import { McpContext, McpNamespace } from '@node-in-layers/mcp-server'
import { Config } from '@node-in-layers/core'
import { YourFeaturesLayer } from './features.js'
const create = (context: McpContext<Config, YourFeaturesLayer>) => {
// Adds your tool.
context.mcp[McpNamespace].addTool({
name: 'my-hello-world-tool',
description: 'My Tool',
execute: async (input: any) => {
return 'Hello, world!'
},
})
// Create a tool from your feature
context.mcp[McpNamespace].addTool({
name: 'my-hello-world-tool',
description: 'My Tool',
inputSchema: {
type: 'object',
properties: {
name: {
type: 'string',
},
},
required: ['name'],
},
execute: (input: any) => {
// You get an object, pass it back to your feature. Handles async for you.
return context.features.yourDomain.yourFeature(input)
},
})
return {}
}
export { create }
Adding Models
You can wrap your models with CRUDS functions and add them to the MCP server with the mcp layer.
NOTE: In order for this to work your layer must have both a services and a features layer. (In addition to your models.) Node in layers will automatically create a cruds property for you with your models, and you can add them.
Here is an example of doing it one at a time. (Not generally recommended, but doable).
// /src/yourDomain/mcp.ts
import { McpContext, McpNamespace } from '@node-in-layers/mcp-server'
import { Config } from '@node-in-layers/core'
import { YourFeaturesLayer } from './features.js'
const create = (context: McpContext<Config, YourFeaturesLayer>) => {
// Adds your models cruds through features.
context.mcp[McpNamespace].addModelCruds(
context.features.yourFeature.cruds.Cars
)
return {}
}
Here is a way that you can really cook with gas. (Highly recommended)
// /src/yourDomain/mcp.ts
import { McpContext, McpNamespace, mcpModels } from '@node-in-layers/mcp-server'
import { Config } from '@node-in-layers/core'
import { YourFeaturesLayer } from './features.js'
const create = (context: McpContext<Config, YourFeaturesLayer>) => {
// This automatically adds ALL of your models from features.
mcpModels('yourDomain')(context)
return {}
}
Another way to organize adding models is from a centralized mcp domain. Put this as your very last domain after all your other domains have been loaded.
// /src/mcp/mcp.ts
import { McpContext, McpNamespace, mcpModels } from '@node-in-layers/mcp-server'
import { Config } from '@node-in-layers/core'
const create = (context: McpContext<Config>) => {
// Add all your models for your whole system in one go.
mcpModels('yourDomain')(context)
mcpModels('yourDomain2')(context)
mcpModels('yourDomain3')(context)
mcpModels('yourDomain4')(context)
return {}
}