This guide is the fastest path to a useful first launch in drio. It is written for GTM and marketing teams who want to ship something real before investing in deeper technical customization.

Before You Start

Come in with these four things:

one clear use case, such as product discovery, support deflection, or lead qualification
one source of truth, such as a public API, product catalog, help center, or internal service
three to five real questions your audience will ask
a rough sense of the brand experience you want the app to match

What You Will Build

By the end of this guide, you should have:

one app with a clear name and purpose
two to five tools that cover the main user requests
widgets that make answers easy to browse in chat
a branded experience you can review internally
a live MCP endpoint you can test in an AI client

Build It Step By Step

Create the app around one audience

Start with a single audience and a single job to be done. Good first examples:

“Help prospects choose the right plan”
“Answer common implementation questions”
“Surface campaign assets for sales teams”

Give the app a name your team will immediately understand.

Add a small set of focused tools

Think of a tool as one thing the assistant can do well. Most first launches only need a handful:

search products
compare plans
answer FAQs
collect contact details
show next steps

If a tool feels like “do everything,” it is too broad.

Connect a reliable data source or API

In the builder, define how each tool gets its information. Start with the cleanest source available instead of the most ambitious one.

Reliability matters more than completeness in the first version. A narrow, trustworthy app is better than a wide, inconsistent one.

Choose widgets that help users act quickly

Match the response format to the user’s next decision:

use cards or carousels for browsing options
use tables for comparison
use markdown or accordions for explanations
use forms when the assistant should collect structured input

If the response is hard to scan, the experience will feel weak even when the data is correct.

Brand, test, and publish

Apply your logo, colors, and copy style. Then test the app with real prompts from your audience, not idealized demo prompts.

When the app feels stable, copy the MCP endpoint and test it in the AI client your team actually plans to use.

Success Checklist

the app answers the top three questions you planned for
the tool descriptions are specific enough that the model uses them consistently
the output format helps users decide what to do next
the app feels branded enough to share internally
the live endpoint works in your target AI client

When To Bring In A Technical Teammate

You can stay mostly in the dashboard for many first launches. Bring in a technical reviewer when:

the app needs per-user access to tools like Salesforce, HubSpot, or internal systems
the upstream API requires non-trivial auth or request shaping
the team wants direct management API automation
security, logging, or deployment questions need sign-off

Launch Your First App