I just built this AI voice receptionist

that can do pretty much anything. All of

the backend automations are happening

thanks to this Naden MCP server that has

seven different custom workflows that I

built out in NAND. So, in today's video,

I'm going to be showing you guys how I

configured Vappy on the front end and

Naden on the back end and more

importantly showing you guys how you can

replicate or customize this system. Not

only am I giving you guys the system

prompt and all the workflows that you'll

see today for completely free, I'm also

chucking in this 15-page resource guide

that walks through the entire

configuration and pretty much everything

that you need to know about this system.

So let's hop into a live demo and then

let's walk through the build. All right,

so on the lefth hand side we have Vappy.

This is Kylie, our Hercules detailing AI

receptionist. And on the right hand side

we have our MCP server with these seven

custom workflows that we built out in

Naden. So before we give Kylie a call,

I'm going to pull up the calendar on the

lefth hand side and the CRM on the right

hand side so that we can see all of this

updating and happening in real time.

We're going to do two scenarios. I'm

going to call in as a new user and have

an account created and book an

appointment. And then I'm going to call

back and I'm going to ask if I can

change that appointment. All right. So,

we're gonna go ahead and give Kylie a

ring and we'll see how this works.

>> Hi, this is Kylie with Hercules

Detailing. Can I please have your email

so I can look you up in the system?

>> Yeah, I'm actually a new caller, so I

won't be in the system.

>> Welcome. Super glad you called Hercules

Detailing today. Uh, all I need is your

email address, your full name, and phone

number to get you set up. Just let me

know those details. Don't worry, I'll

confirm them with you so we get

everything perfect. Go ahead whenever

you're ready.

>> Sure. Yeah, my email is nateample.com.

My name is Nate Herk. That's spelled

with a K at the end. And then my phone

number is 3334445678.

>> Awesome. Thanks for that, Nate. Just to

confirm,

your email is nateample.com.

Your full name is Natek with a K at the

end and phone number is 3334445678.

Right? If that all looks good to you,

give me one second to send that in and

get your profile set up.

>> All set, Nate. You're officially in our

system. Super easy. Now, what can I help

you with today? Are you looking to book

an appointment? Have a question. Or is

there anything else on your mind?

>> Yeah, I'm just looking to schedule an

interior detailing for tomorrow.

>> Great choice. Interior detailing will

have your car feeling fresh. Just give

me one second and I'll check our

availability for tomorrow real quick and

let you know what times are open.

>> Sounds good.

>> Okay, here's the scoop.

All right, here's what tomorrow looks

like. We do have some time slots already

taken. One in the morning from 10:15 to

11:15 and another from 2 to 3, but every

other hourlong slot during the day is

open. Just letting you know what time

you'd like to come in for your interior

detailing. Appointments last one hour.

So you can pick any time slot outside of

the 10:15

11:15

and two three windows. What works best

for you?

>> Yeah, let's do an 8 a.m. slot.

>> Give me one second to process that for

you.

All right, Nate, you're all set for an

interior detailing appointment tomorrow

at 8 a.m. Your spot is reserved and

ready to go. If you need to change

anything or have any questions about

your appointment or our services,

seriously, just let me know.

>> Anything else you'd like help with?

>> Nope, that's it. Thank you so much.

>> Sweet.

>> All right, so just had to hang up on her

there. So, we can see that all that

updated in real time. And if I now go to

the call log, if we just give this a

sec, this should trigger over a summary

of the call, as you can see, as well as

the outcome, which was appointment got

booked. And the summary was that we set

up an account. And now we've booked in

an appointment for 8 a.m. So now we're

going to run another example where I

call in and I ask if I can move this

appointment. So I'm going to go ahead

and give Kylie another ring.

>> Hi, this is Kylie with Hercules

Detailing. Can I please have your email

so I can look you up in the system?

Yeah, it's nateample.com.

>> Awesome. Thanks, Nate. Let me check on

that real quick and see what we've got

for you in the system. Just give me a

sec. All right, Nate. Looks like I found

you. Welcome back. What can I help with

today?

>> Yeah, I just want to actually move my

appointment. I was going to see if I

could just bump it an hour later.

>> Cool. No problem at all.

Let me check on that real quick so we

don't double book anyone.

I want to see what's available for you

around that time. Give me just a second.

All right, Nate, here's the scoop. After

bumping your appointment an hour later

to 9:00 a.m., it looks like that time is

totally open for you. No conflicts at

all in the calendar, so you're in the

clear. Um, just to confirm, you're

moving your interior detailing

appointment, right? Ready for me to

update it to 9:00 a.m. for you? Let me

know. and I'll get it all sorted out.

>> Yeah, that's perfect. Let's go ahead and

do it.

>> Give me one sec to process that for you.

>> Boom. It's done. Your interior detailing

appointment is now officially moved to

December 7th at 9:00 a.m. and it'll run

until 10:00 a.m. You're all set.

Anything else you want to update or ask

about while I got you online? Nope,

that's it. Thank you so much for your

help today.

>> You got it, Nate. Thanks for call.

>> All right, hang up on her again. And now

let's just give it one sec and we will

see that that call comes in and it

should say that everything was updated.

As you can see, there we go. We just got

that call log and we can see the outcome

here was that the appointment got

rescheduled. And once again, in the

appointment log, we can see under the

notes that this appointment was moved to

9:00 a.m. instead. All right, so now

that you guys have seen a live demo or

maybe I guess heard a live demo, let's

jump into how I went about building this

thing from completely scratch. All

right, so I'm sitting here editing this

and I just wanted to say something

really quick. The idea of these systems

is to make them as good as possible so

you can make the experience for the

caller as good as possible. But I still

think that the AI systems, whether

that's inbound or outbound, should start

off the conversation by saying, "Hey, I

am Kylie, an AI assistant for Hercules

Detailing." because it'll just never be

perfect, you know, with interruptions or

with like little just kind of etiquette

things. So, I think that you should

always have an AI caller tell the human

that, hey, I'm an AI caller. But

anyways, let's get back to the video.

All right, so in a voice agent, there's

really two main pieces. You've got the

front end, which is like the voice AI,

and you configure it in something like

Vappy, which is what we're doing here

today. And then you have the back end,

which are some of the actual automations

that are taking place. And as you guys

know, we're using these seven edited

workflows that are being communicated to

Vappy with this MCP server trigger. So,

if you've never heard of Vappy before or

the idea of a voice agent is very

foreign to you, then I would definitely

recommend checking out this video, which

is kind of a prerequisite. It's like a

40-minute full beginner's guide, and

then hop into this video. So, I'll tag

that right up there. So, I'm going to

break down with you guys exactly how I

went ahead and built this and how you

should go ahead and think about building

a voice agent system. And it all starts

with some sort of wireframe or flowchart

because there's so much conditional

logic. There's so many different

scenarios that could play out,

especially in a voice conversation. So

having all of this at least roughly

identified as far as if X do Y, if A do

B, and you have all this on paper, it's

going to make your job so much easier

when you get into Vappy and you start

system prompting this thing. So before

we get into Vappy or End, I'm going to

walk you guys through my wireframe. I'm

going to walk you guys through my

thought process with the actual backend

automations and then we're just going to

take it step by step and we're going to

see all of the different scenarios play

out. Okay, so starting up here at the

beginning, we know that what's going to

start the process is an inbound call. So

we have a number on our website or

whatever it is and a user is going to

call the number. Now, I'm not saying

that this is the optimal flow for like

an AI receptionist, but this is just

what I wanted to do. So I wanted to have

Kylie start off by asking them for their

email so she could see if they already

exist in the CRM. And so I knew we'd

have to be doing some sort of CRM tool

to look up the client. From there, I

knew that there were going to be two

paths. They either exist in the database

or they don't yet exist. And so if they

exist, we'll gather personal info. And

if they don't exist, we'll go ahead and

create a CRM entry for them. Now, that

was basically step one. And then after

that, we merge back into the same path,

which is to gather intent. And right

here we can see that we gather intent.

Whether that's something general that

Kylie, the AI receptionist, will be able

to handle or if it's something more

specialized like transferring to sales

or transferring to customer support.

After intent is gathered and if we know

that it's going to stick with Kylie, we

have a few different things that could

happen. So we know that we could have a

caller asking questions. And if that

happens, we need Kylie to answer

questions with real truthful

information. So we're going to give her

a knowledge base, which is a real

document. And that might be a loop of,

you know, five or six questions. The

other scenarios are that they might want

to do something with appointments. So

that could be booking, it could be

changing, or it could be deleting. And

so there's a few dependencies here. If

they want to book an appointment, we

first have to check availability. If

they want to change an appointment, we

first have to look up the appointment to

change it, but also check availability.

If they want to delete an appointment,

we have to look it up first in order to

be able to delete it. And then after all

of these actions have been taken, we

want to use the CRM to actually add

activity back into that log. And so sort

of after this loop of you know intent

being handled then we will go ahead and

end the call and that's where we want to

log the call into our call logger

system. So that's kind of the flow that

we've built out today. And now let's

talk about what I was thinking after I

did this. And I was like okay well let's

actually think about how these tools

will work because VAP is going to have

to pass over certain variables to each

tool and those are all going to be

different. So how do I want to think

about doing that? So my first thought

was I was going to have two editin

workflows. One would be a CRM tool and

one would be an appointment tool. And

each of these would have a web hook with

an intent check at the beginning. So CRM

could be the intent to look up an

appointment or to add activity or to

look up a client or to add a new client.

And then similarly for appointment, the

intent could be checking availability or

to create change or delete an event. And

so the reason I needed to do this was I

needed to start thinking about what

variables will need to be sent over to

each of these kind of like subflows in

order for Vapian to be on the same page.

But as I started to build this out, I

realized I might be over complicating

things because with Vapian Nadn, when

you have it sending to a web hook, you

have to respond to that web hook. And so

it just became complicated. And I

realized that what I could do is just

have an MCP server in NAND that has

access to all the correct workflows and

it would be taken care of really

quickly. So I know that a lot of this

may just seem confusing because it's all

just little colored boxes and squares

and rectangles. So let's start to get

into Vappy and Niten and let me show you

guys how all of this stuff actually

works. Okay, so here we are in Vappy. We

have our model, which is kind of the

first thing, which is OpenAI GBT 4.1. I

found that 4.1 was a lot smarter and

better at following instructions than 4

was. We've got the first message, which

is Kylie saying, "Hey, this is Kylie.

Can I please have your email so I can

look you up in the system?" We've got

the system prompt, which the hard thing

about system prompts with the voice AI

is that it's just prompting and testing

and prompting and testing, and that's

really the only way to do it. And this

is probably the 50th version of the

system prompt that I've done. But the

cool thing about Vap and system

prompting is when you click on generate,

you can actually have it edit certain

things in your prompt. And it makes it

really easy to iterate upon your current

version. Because if I say something like

make Kylie super sarcastic, it'll go

ahead and edit my entire system prompt.

And then it will also show me exactly

where it made changes. So you can see

what it crossed out and what it added.

And then I can go ahead and either

refine those edits or I can just accept

the changes or discard. So I'm just

going to go ahead and discard because

that was just to show you guys what that

looks like. So we talked about how we

have the front end with Vappy and we

have the back end with Naden. So how do

we actually connect these two things?

Well in Vapy we have a tool for our AI

assistant called NADN. And you can also

see in our system prompt there are

instances where we say something like

use the NIAN tool to check the CRM. Use

the NADN tool to check for availability.

And so that NIDAN tool looks like this.

It is our MCP server with all seven of

these workflows that we also built out

in NIDN. And so I know this seems like a

lot. We're going to run through all of

it, don't worry. But these seven

workflows, I just wanted to real quick

show you what they look like. So, here

is the client lookup tool. Here is the

new client CRM. Here is the check

availability. Here is the book event.

Here is the update appointment. Here is

the lookup appointment. And then here is

our delete appointment. And what you'll

probably notice as I'm flipping through

a lot of these tools is some of these

are very simple. Like this one's

literally just one node and this one's

two and this one's one. This is the only

one that really kind of has like

multiple conditional checks. But that's

what's so cool about this is that each

tool has one very specific function. And

so it's just basically up to Vappy and

the MCP server to figure out which one

to call and then it can get answers back

so quick. Because one thing I see a lot

that really kind of makes me cringe when

people are building voice agents and

using NN is building an NN AI agent in

the back end to actually make decisions

and to do things because what you're

doing there is you're doubling the

amount of reasoning and cost and latency

because what happens is Vappy right here

this entire system prompt this is an AI

agent in itself because it has to use a

brain to understand what to do. And so

if you use a brain to understand what to

do to trigger another brain that needs

to understand what to do, then you're

just like, like I said, you're just

doubling the amount of reasoning and

potential errors. So by having Vappy

control everything very clearly and then

having that send over to an MCP server

that just makes one very quick request

and we keep it on the guard rails

because it's just a very standard

workflow. You'll notice in all of these

backend workflows, there's no AI going

on at all. That's how we can really keep

these voice agents running fast and

doing what we actually want them to do.

But anyways, there's so much to dive

into here. So, let's go ahead and talk

to the assistant and just play out that

first scenario, which is creating a new

account. And I'll show you guys how that

actually works in Naden. And I obviously

could call her on my phone, but I'm just

going to talk to her in this environment

so we can all see the transcript and

everything. It makes it a lot easier.

All right. So, you can see it called the

client lookup tool in NAN. And then it

called the new client tool in NAND. So,

let's go ahead and hop over real quick

to Vappy. You can see that if I click on

executions here, we have two new ones.

So the first action that it took was the

lookup. And if I click into here and I

go to the view subexecution, you can see

that what happened here was it sent over

mike@dm.com.

It looked him up in the CRM and it

realized and it realized that there's no

record for Mike. So nothing was output

which is why it gets sent down this next

path which basically filters it down

this way and then it sends back this

message to Vappy which says this is a

new client. They don't exist in the CRM.

So after Vappy saw that and confirmed

all of our details, it went ahead and

triggered off the new client CRM tool.

So when I click into this subexecution,

you can see that Vappy sent all of these

details to NADN and then NADN went ahead

and logged it in the CRM and that is how

we got this new entry right here. So I

know you guys are probably wondering how

does the server trigger actually know

which of the tools to call for the given

scenario. Well, Vappy will send over

some sort of context and some sort of

information, but then in all of the

tools, what you can see is we have a

really easy description that shows when

do you call each tool and what do you

do? And it's able to use all of this

context to understand which of these is

actually the best tool for my given

scenario. And so all of the stuff going

on as far as like having them say

something before they call a tool and

confirming all these details, of course,

all of that is in the system prompt and

that's going to be attached in here. The

full system prompt that I'm using right

now in this video is attached in this

document. So you guys can go ahead and

read through it and play around with it

if you want. So I know it may feel like

we're going through this stuff kind of

fast. So if you guys want to see the

full live step-by-step build, which was

like 2 hours of me actually building the

system, then I drop stuff like this all

the time in my plus group, which you can

check out down below, link in the

description. So the question is, how did

Vapy actually call NADN? Well, we had to

give it access to a tool. So you can see

if I click on tools right here, you can

see that one of the tools that we have

is called NADN. And on this left-h hand

side, if I click on tools, you can see

that this is the Naden tool that I

actually configured. So the way that you

can do this is you click on create tool,

you go to MCP as far as the integration,

and then it pulls up a screen like this

for you to configure. I went ahead and

called my tool NN and then it asks for a

server URL. So, if I go back into the

edit and flow with our MCP server

trigger, basically I clicked into this.

I grabbed the production URL and I copy

and pasted this right here into the

server URL. Now, the last thing you have

to do to actually make this work is you

have to send over an API key from NIN.

So, in the headers, you're going to add

a header right here. You're going to put

the name as authorization and for the

actual value, you're going to do capital

B bearer space and then paste in your

Naden API key. And the way that you go

ahead and get your API key from nitn is

by going to your settings which will be

in the bottom left. And then you click

on nitn API. And then you should be able

to create a key right here. And that

value that you get is what you paste in

here after the space. And then you just

want to go ahead and switch it from

streamable HTTP to server sent events.

And then this basically gives the

connection for Vappy to call this tool

and it will be able to look through and

use all of the workflows that you hook

up to that server trigger. But of

course, what you cannot forget to do is

tell it in the system prompt when and

how to use that tool. So right here you

can see we have our identity style

response guidelines. But then when we

move into more of the conditional logic

flow, we have tasks and goals. So we

have the initial greeting and you can

see it says greet the caller warmly and

ask for their email to look up their

profile using the NAND tool. And so

that's how it knows to call it and to

send over their email. I said convert

the email to lowercase before using it

in the CRM lookup with the NAND tool. I

also made sure that I said before

calling the tool you must say something

like let me check on that real quick to

avoid silences because otherwise they're

just going to call the tool and there's

going to be like 5 seconds of silence

while they're waiting. And then moving

on to number two you can see the CRM

lookup logic. I said use the end tool to

check the CRM. Before you call the tool

you must say something. If the client

exists you do this. So you acknowledge

them. If the client's new basically say

that we're going to set you up an

account. We're going to use the end

tool. Make sure to confirm the spelling

of their name before logging it. confirm

the information is correct before you

call the tool. And so that's as far as

we just saw in the demo. So now we have

intent gathering in action and

appointment management and a few other

things later. So let's go ahead and do

another example and we're going to book

an appointment and then I'll walk

through how all of that worked. Okay, so

let's walk through the edit in workflows

that just happened again. So we called

in as mikedm.com and it did a client

lookup but this time it was successful.

So let's go see what that looked like in

edit in. So here was the first execution

where the MCP server trigger called the

client lookup tool. You can see that it

sent over miked dm and this is what that

subexecution looks like. So it looked up

the email mike@dm.com and it came back

with a record which means we send it

down the true branch which basically

says that they already exist in the CRM

and from there we go ahead and look up

their email in the CRM to see if they

have any appointments and currently they

do not have any appointments in the log

which is why it goes down this false

branch which basically says the client

exists however he or she has no previous

appointments. And if they did have

appointments, it would go up this way

and then Vappy would be able to see

that. Now, after that, what happened in

Vappy is we looked to book an

appointment. So, first Kylie had to look

up and check availability and then she

had to confirm the time and then go book

the event. So, let's go take a look at

those two tool calls. So, the first one

was checking availability. It called the

check availability tool. And here's what

that subexcription looks like where it

basically sends over the current time

and then the end of the day. So, it said

12:08, which is right now when I made

the call. And then it's saying what

times between now and the end of today

are available. And that's what we go

ahead and search in the get many events.

So you can see that there are no events

from now until the end of the day which

is why nothing returns. And the if node

is checking if anything comes back

because if there are events it will

basically send it up the true branch and

then Vappy will see the booked times.

But because there's nothing it comes

here and then it says the entire day is

available. And then of course once the

time has been confirmed we go ahead and

call this book event tool which sends

over the start time and the end time.

the email and the actual event summary

for the event to book. And that

subexecution is very simple. We're

basically passing that information into

calendar to create an event. And then

we're passing it into our Google sheet

to create a new row in the appointment

log. So now if I was to switch over to

the CRM and click on the appointment

log, we can see that we just got a new

appointment booked here with the event

ID, the email, the type, and the date,

as well as the notes saying that

appointment has been booked. So now to

put that all together, the tools have

been called, the things have happened,

but how did we actually get that to

happen in the system prompt? So what we

did is we told it exactly what to do for

each scenario. And right here, we

skipped over the intent gathering as far

as like transferring or handing off. So

I'll do that example next. But for

appointment management, what we said is

for all of these actions, you first have

to check availability. You're going to

call the end tool to assess appointment

availability. If it returns the entire

day is available, inform the client

accordingly. If it provides specific

time slots, then just know that those

are busy times. If checking availability

for today, then send over the current

time up until 235959,

which as you can see in the check

availability tool, it sent over 1208,

which was the current time when I made

that call. And that's how it knew that

there are no events from now until the

end of the day. And apparently Hercules

Detailing is open 24 hours. But then we

also said if checking for a specific

date, so for like tomorrow or days in

the future, use a 24-hour format and

check from the beginning of the day all

the way up until the end of the day.

Only tell the caller what other times

are available. Don't tell them what

they're titled. After a time has been

confirmed, inform the client that

appointments last 1 hour. And so, as you

can see here, I'm just basically giving

it really, really specific guidelines as

far as like follow this script. So, if

you think of like an actual call center

where humans are making the calls, they

typically have a a script with all this

conditional logic of what to say and

when. And that's really what we're doing

here. So, let's real quick go ahead and

do a quick example with updating or

deleting an appointment. And then we'll

jump back to how you can transfer or

handoff calls. Okay. So, I just finished

talking to Kylie. She started by looking

us up in the CRM. I said that I wanted

to push the appointment back 1 hour. So,

first she checked on the availability.

She then said that okay, that's open.

So, then she wanted to look up our

appointment to actually move it. She

called me buddy, which I didn't love,

but she said that it is available. And

then when I confirmed I want to move it

back an hour, she went ahead and used

the update appointment tool. So, let's

go take a look at Niten and what it

actually did there. All right. So, here

are the four executions that it had.

We're not going to look at the first one

cuz it's just a client lookup again. The

second one is where it went ahead and

checked availability for the rest of

today. And then if we go to this one

which is the lookup appointment, let's

see what's going on in this

subexecution. So here what we see is

Vappy sent over the after time and the

before time. And what that does is it

gets any events that are within that

window. And so it comes back with our

event because what it's looking for is

this event ID so that it can actually go

ahead and update it later. So it takes

that event ID that we just saw and then

it calls the update appointment tool.

And you can see that's what it passed

over right here as well as the new start

time and end time. So let's click into

the subexecution here. So it's really

pretty simple. It takes that

information. It creates that new

calendar event and then it goes ahead

and updates that row in the Google sheet

because we're able to match on the event

ID which is a unique identifier. And

that's why if you remember in our

appointment log, this originally said

appointment booked and now it updated

the date and it also changed the notes

to saying we had this rescheduled or

this one was moved. And this is what the

delete appointment workflow looks like,

which is the exact same except for

instead of updating an event, we're just

deleting it. So it would basically send

over the event ID and then we would

delete it through that. And then we

would do the same thing in this Google

sheet where we're matching on the event

ID. And then we're just changing the

notes to say cancelled. Okay. So in our

wireframe, we've seen all of this. We've

seen the different paths. We've seen

pretty much all of these different

options for booking appointments and the

CRM tool. But what we haven't yet seen

is transferring calls and the knowledge

base. So, let's first do this and then

we'll go ahead and take a look at this.

>> Hi, this is Kylie with Hercules

Detailing. Can I please have your email

so I can look you up in the system?

>> Nope. I don't want to give you my email.

I just want to talk to the customer

support department because I'm furious.

>> Totally get it. If you'd rather not

share your email, that's all good. Would

you like me to transfer you to our

customer support team right now so you

can chat with them directly?

>> Yes, please.

I think you'll

>> Hello, I am customer support.

>> Okay, so as you can see right there, we

had a angry customer. They didn't want

to give their email, but that was fine.

Kylie just basically said, "Okay, I'll

transfer you." And then we used a tool

called Handoff, which if you remember,

and I go to the tools that this

assistant has access to. The other one

that it has access to is called Handoff.

So the way that we actually configured

this was back in the lefth hand side.

We're going to click on tools, and you

can see that we have a tool called

handoff. So when you go ahead and create

a new one, you see that you have

transfer call. But what transfer call

lets you do is this lets you actually

put in a phone number if you wanted to

transfer this to a human. So you could

add a destination and put in a phone

number. But when you do handoff, what

this lets you do is it lets you hand it

off to a different internal Vappy

Assistant. So you can see here that what

I'm doing is I added a destination to

send it over to a Vappy Assistant and I

was able to choose that I wanted to send

it to our customer support specialized

agent. And what's cool is you can have

multiple different destinations within

one handoff tool. So this handoff tool

could send one to customer support, it

could send a different path to sales,

another one to Alex. And basically when

Vappy calls this tool, it figures out

which of the ones to hand off to. And

it's able to understand that because

right here it says hand off to this

assistant when the caller requests

customer support or has a customer

support related inquiry. And so

orchestrating all of these different AI

systems together, it's all about

context. So this is basically how it

knows to hand off to customer support.

And if I go back into our MCP in Nitn,

you can see that each of these different

tools has a description. So that's how

the server knows based on the request

coming in, which one of these tools am I

going to call? Okay. So the final thing

that we have yet to look at is the

knowledge base and the ability for Kylie

to just answer general questions. So

let's say you have a document like this,

which is Hercules detailing, policies,

and FAQs. You've got stuff like business

hours, location, contact, email, all of

this kind of stuff. And so what you do

is you can take this as a PDF. You can

go into Vappy and you can go down here

to more and click on files and you can

upload a file right here. So you can see

that I dropped in the Hercules detailing

policies and FAQ. And now if I go back

into our actual assistant, the AI

receptionist at the end of the model,

you can see that we have the ability to

add files. And I've added the Hercules

detailing policies in FAQ. And then what

else I do is in here in the system

prompt right here for general question

handling. I said for general questions

about location, business hours, general

policies, FAQs, use only the information

from the Hercules detailing policies and

FAQ using the default query tool. Do not

make up any of your information. And so

it's interesting because the default

query tool just means that it's looking

at this file even though you don't give

it access to a tool called default query

tool. So, let's go ahead and ask a

question where it would have to use that

file. And you'll see what I mean in this

little interface over here how it calls

the default query tool.

>> Can I please have your email so I can

look you up in the system?

>> Yeah, I don't really want to provide my

email right now. I just want to ask

about the different services that you

offer if you have like different tiers

or what else you guys offer just besides

a detailing.

>> No worries at all. You're totally

welcome to just ask questions. Um, let

the info on and I'll make it um super

easy. I promise. Just give me a sec. All

right, here's a quick scoop. We got

three names of that we need. One,

>> okay, so I had to cut her off, but you

can see that she was able to use the

default query tool and was able to

answer us with all of this information

because we gave her a source of truth

file. Okay, so now that you've seen this

entire flow, before we actually dive

into the log call operation, I just

wanted to zoom back in to why I actually

made the decision to switch from two

workflows to an MCP server. because now

we have like a 50,000 foot view of how

this works. So if I wanted to send data

and have different tools for each of the

different like operations, we would have

to do a lot more prompting because we'd

have to specify in this case use this

tool, in this case use that tool. But

also, if we wanted to add those as

tools, we would basically be doing this

custom tool operation and we'd be naming

it. We'd be giving it a description. We

would be configuring the web hook

address to send it over to, but we would

also be defining the parameters. So if

you remember when we want to do

something like create an appointment we

need email time and summary but if we

want to change it we have to get e event

ID email time and summary and if we want

to delete it we just send over event ID

and so there's all these different

parameters that we would have to

configure and say hey Vappy when you

call this tool send over these three

things when you call this tool send over

those five things and that means if we

ever wanted to change something in the

backend or you know whatever other

scenario we would have to make updates

in two spots.

But because we have an MCP server where

all we'd have to do is configure how

these workflows take inputs, the server

trigger basically is able to find all of

the inputs and the schemas and just send

over what we need because Vappy

understands the use case in our

assistant once again because we prompted

it right with everything that it needs

to know. So hopefully that all makes

more sense now that we've seen all the

tools in action. All right. So, now that

we've gone through this entire flow and

seen some different scenarios, how does

it actually log this information in our

call logger? Because remember in our

CRM, we have clients appointment log and

then we also have a call log where we

can see all of the different times of

the calls and the summaries and the

outcomes. So, first of all, in our AI

receptionist, I could click on these

three dots and I could go to call logs,

which is great for visibility and for

being able to do a lot of testing

because it's going to let you actually

listen to all of the different calls and

you can see what happened in them and

like what tools they called. But the way

that you can get all that detail sent to

some sort of like actual external logger

is if we go to the advanced section and

we go all the way down here to

messaging. And if I open up messaging,

you can see that there is something

called server settings. And this is

where I gave it an edit in web hook to

log our information. So this is the web

hook. You can see that it's active. So

what I did is I opened up the web hook

and I grabbed the production URL. Even

though when you click into it, it looks

like it's open on the test, but if it's

active, it will be pulling the

production. So I grabbed this. I copied

this right here into our server

settings. And then if I scroll down a

little bit more, you can see that we

have client messages and we have server

messages. So right now I'm just worried

about server messages because it says

these are messages that will be sent to

the server URL configured which is what

we just put up here which is our NN web

hook. And so in here by default there

will be some more things selected but

all I wanted to do was end of call

report because you can also see that

other things can be sent over like the

transcript like the tool calls things

like that. But the end of call report is

going to give us these things up above

that we configure. It's going to be in

the analysis section. It'll give us the

summary and it will give us any of this

structured data that we want it to send

over. So for the summary, it's really

simple. I said you are an expert

notetaker. Your job is to summarize the

transcript in two to three sentences. So

you can customize what type of summary

you want. But if you're actually looking

for key pieces of information, you can

use structured data to basically like

pluck out certain things that you want.

So in our case, we're only sending over

an outcome as you can see, which is like

one of these types of things. But maybe

in your call logging, you also want to

get plucked out like the email of the

person or the type of appointment they

booked or the time of the appointment

they booked. And that's why we have all

of that being logged already in the

appointment log. But you could do that

as well here in the call log because all

you'd have to do is basically tell it, I

need you to pick out these different

variables and then send them over in

these different properties. And you

could add as many as you want here. And

so this stuff will automatically be sent

to the correct URL if you fill one out

here. And so if I go into Hercules

detailing receptionist end of call

report and I click on executions, what

that's going to do is it's going to show

us the payload that Vappy sends over

here. So with that most recent call, if

I click on the payload and we see all of

this nasty stuff, but what I can do is

just go to schema and what I want to

look at is the body. And this is where

we're getting our summary sent over as

well as the structured data that we

configured which is called outcome. So

the summary is that the customer called

Hercules Detailing to inquire about

services in different tiers. the AI

began explaining the three main tiers

and the outcome was that customer

inquired about services. So really like

the same way that you want to have a log

for your automations to see what

happened and what was called, you can do

that with your voice agents as well. And

basically we just write that into our

Google sheet with the following three

fields. So the last thing I wanted to

talk about was how I connected this to a

phone number. So, in Vappy, if you want

to create a phone number, you can grab a

free Vappy one, and it's up to 10 per

account, which is great. And this is a

US phone number, but you can also import

other ones if you already have phone

numbers that you want to hook up. Now,

once you have a phone number in Vapy,

it's up to you to assign it to

something. So, in here, all I did was I

assigned it to an inbound phone number

and I plugged it in to the AI

receptionist. You can see that when you

do that, you could add a fallback

destination or you could also make it an

outbound phone number if you wanted to

have different assistants calling

outbound. If you guys want to see a

tutorial of an outbound voice agent,

then let me know some use cases and I

will definitely make a tutorial on that

for you guys. But I've got a feeling

that this video is running pretty long

as is and I don't want to make it any

longer. So, what I have to wrap up with

is just telling you guys how you can

access all of these workflows as well as

this document that has all the workflows

and the system prompts in it and

everything like that. So, you're going

to join my free school community. The

link for that is down in the

description. And once you get in here,

all you have to do is go to YouTube

resources or search for the title of the

video or you can go to free nad

templates. And once you get to a post

associated with the video, let's say

it's this one. You will be able to see

right here the PDF will be attached as

well as the JSON or in this case it'll

be maybe a zip file with a bunch of

different JSON workflows. And then you

can just basically configure everything,

point it to Vappy, and start testing it

all out. And if you want to access

unlimited support and connect with over

200 members who are building with NNN

every day and building businesses with

NNN every day, then definitely check out

my plus community. The link for that is

also down in the description. Like I

said earlier, I posted the full live

2-hour step-by-step build of me making

this AI voice receptionist system in

this community already. And we also have

a full classroom section with tons of

other projects that are live. We have

agent zero, which is the foundations for

beginners. We've got 10 hours and 10

seconds where you learn how to identify,

design, and build time-saving

automations. We've got one person AI

agency and we've got subs to sales. So,

I'd love to see you guys in this

community in the live calls. But that's

going to do it for today. So, I hope

that you guys enjoyed the video. If you

did, please give it a like. It

definitely helps me out a ton. And as

always, I appreciate you all making it

to the end of the video. I'll see you on

the next one. Thanks everyone.