Model Context Protocol
From start to finish, set up an NLX application exposed through Model Context Protocol (MCP)
Last updated
From start to finish, set up an NLX application exposed through Model Context Protocol (MCP)
Last updated
Model Context Protocol (MCP) is a standardized protocol for how Large Language Models (LLMs) integrate and engage with external systems. Prior to MCP, builders would need to provide a multitude of tools via REST APIs to give an LLM the ability to perform even the simplest of tasks you wish to define:
Check the weather local to a user
Provide an activity recommendation local to a user
NLX’s MCP support allows you to turn any NLX application into an MCP Server, giving an LLM (MCP Client) the ability to follow and perform tasks outlined in the application's as well as easily to NLX when the LLM interfaces with a user.
Your NLX flows effectively become MCP tools that you provide to a supporting LLM to boost its capabilities. This process is done entirely without writing code or exposing your systems and services to new vulnerabilities.
You'll complete the following to successfully launch your MCP implementation:
Est. time to complete: ~10 minutes
Each flow is invoked when your chosen AI model identifies customer intent from a user's query ("What's the weather like?") and matches it to a flow you've created (WeatherUpdate).
Variables required for the flow to work that should be extracted by the LLM in conversation and passed along to NLX are set up first in the flow's Settings:
Select Flows in workspace menu > Choose New flow > Enter a descriptive name (no spaces or special characters) > Select Save
Choose Settings (gear icon) in flow toolbar
From the AI settings tab:
In the AI description field, enter a concise description explaining the purpose of the flow that LLM models reference to invoke the flow
Enable MCP toggle
Provide a unique and concise input name (no spaces or special characters)
Expand each property variable defined in your MCP input schema > Expand their settings
Enter a brief description in the property's Description field for the LLM to understand the purpose and context of each (e.g., location
property might have the accompanying description for a weather update flow: The location of the weather request
)
Click Save
On any node of the flow, enter an open curly brace {
and reference the MCP input variable you want to use as an output in messaging, payload fields, Split node conditions, etc:
Est. time to complete: ~10 minutes
Now you'll set up and deploy an application to become your MCP server.
Select Applications from workspace menu > Choose New application
Enter a descriptive name > Click Save
Click Flows tab of application > Select Attach flows > Attach one or more flows created to make available to your application > Click Attach selected
Select Channels tab of application > Expand API option > Click + Create channel
Enable MCP interface toggle
Click Create channel
Select Settings tab of application > Under AI settings, enter a concise description of the application's purpose into the AI description field
Click Save
A build constructs the array of flows that make up your conversational AI application and updates any changes made to your flows, while deploying makes a successful build live:
Click Deployment tab of application > Select Create or Review & build
Wait for validation to complete > Select Create build*
When satisfied with a successful build, click Deploy
Est. time to complete: ~5 minutes
To make your application available to a supported MCP Client (Large Language Model service), complete the following:
From the Deployment tab of your NLX application, select Details next to the Deployed status
Expand the API section under Setup instructions in the pop-up > Copy the MCP URL
and the API key
Use the following for the claude_desktop_config.json
and replace with the MCP URL
and API key
copied earlier:
Begin by identifying the tasks your conversational AI application will automate and organize them into individual topics handled by . Determine the sequence of steps and messaging that the conversational application follows to assist a user with the task. The conversation workflow is assembled in a flow's Canvas with a pattern of nodes similar to a flow diagram.
Enter the input schema containing that will be set and passed along by the LLM interfacing with a user
*After a build status appears as Built, you may use the Test feature to test the conversation with your application using the latest build.
Looking for more? See
Complete the MCP setup for your preferred MCP-supporting client. For example, here are the