1
00:00:00,200 --> 00:00:03,320
All right, so at the beginning 
of April I went to New York for 

2
00:00:03,320 --> 00:00:07,400
the MCP Dev Summit, and while I 
was there I got the chance to 

3
00:00:07,400 --> 00:00:11,080
record a few podcasts with 
attendees that were at the 

4
00:00:11,080 --> 00:00:12,640
event. 
This is one of those 

5
00:00:12,640 --> 00:00:14,120
conversations. 
Hope you enjoy. 

6
00:00:14,320 --> 00:00:16,480
I think the main thing is that 
people don't quite understand 

7
00:00:16,480 --> 00:00:18,760
what it is that easily. 
People think it's some way of 

8
00:00:18,760 --> 00:00:20,960
mixing AI into your API in some 
way. 

9
00:00:20,960 --> 00:00:23,440
Like API is going to change what
it does and magically it's going

10
00:00:23,440 --> 00:00:26,360
to sort of do new things, which 
of course isn't what it is. 

11
00:00:26,600 --> 00:00:30,240
It's AI using AP is there's a 
lot of questions you get that I 

12
00:00:30,240 --> 00:00:32,720
can't answer that well because 
they're about other ways to 

13
00:00:32,720 --> 00:00:35,640
actually like inject prompts 
back into the LM or you know, 

14
00:00:35,640 --> 00:00:37,760
other things. 
But I think part of it is just 

15
00:00:37,760 --> 00:00:41,040
that if you're using MCP to use 
your API, it's good to not 

16
00:00:41,040 --> 00:00:43,560
exposed to much of it, and it's 
good to tell us if API be very 

17
00:00:43,560 --> 00:00:57,840
secure in the first place. 
Hi, I'm Alex Saltkeever and I'm 

18
00:00:57,840 --> 00:01:01,240
here at MCP Dev Summit North 
America in New York City. 

19
00:01:01,600 --> 00:01:06,640
I'm here with Ben Morse of Deep 
Bell, which is a Tell us what it

20
00:01:06,640 --> 00:01:08,000
is, please. 
What is Deepel? 

21
00:01:08,000 --> 00:01:12,560
Deepel is AAI language 
translation company based in 

22
00:01:12,560 --> 00:01:15,240
Germany. 
We've got about 1000 employees, 

23
00:01:15,400 --> 00:01:18,760
so it's like a popular like 
phone app and web app and such 

24
00:01:18,760 --> 00:01:22,120
thing in countries like Germany 
and other European countries and

25
00:01:22,120 --> 00:01:26,040
also Japan and even Korea not as
well known in North America. 

26
00:01:26,040 --> 00:01:28,840
So hire people like me to be 
here in North America and say 

27
00:01:28,840 --> 00:01:30,640
hey, we exist. 
Hello. 

28
00:01:30,640 --> 00:01:32,680
Hi, here we. 
Are thank you. 

29
00:01:33,040 --> 00:01:36,160
The reason I wanted to talk to 
you though is your recent MCP 

30
00:01:36,160 --> 00:01:39,640
roadshow. 10 countries in 10 
days I believe it was. 

31
00:01:39,680 --> 00:01:42,280
It wasn't quite 10 countries, 
but there were 10 days. 

32
00:01:42,280 --> 00:01:42,960
But there were. 
How did? 

33
00:01:42,960 --> 00:01:46,640
This How did this come about? 
This mad plan to spread MCP to 

34
00:01:46,640 --> 00:01:48,560
the world. 
Yeah, it was kind of, it just 

35
00:01:48,560 --> 00:01:51,440
kind of grew organically. 
So when I got to Deep L and 

36
00:01:51,440 --> 00:01:55,800
became the developer advocate, a
person had this remit to make it

37
00:01:55,800 --> 00:01:59,000
to people knew what Deep L was 
in North America, and the 

38
00:01:59,000 --> 00:02:01,840
developers knew we existed. 
So I had to go out there in 

39
00:02:01,840 --> 00:02:03,880
communities and just find 
people. 

40
00:02:04,120 --> 00:02:06,560
And so I was looking for ways to
do that. 

41
00:02:06,800 --> 00:02:08,800
I thought, here's a hackathon in
San Francisco. 

42
00:02:08,800 --> 00:02:13,000
Like this was like May 2nd. 
In May, Sorry. 10 cities, yeah, 

43
00:02:13,080 --> 00:02:15,640
it was 10 planets and 10 light 
years. 

44
00:02:15,840 --> 00:02:19,400
That was, it was 10 cities. 
I think 10 talks in four 

45
00:02:19,400 --> 00:02:22,160
countries. 
I think, yeah, we're going to go

46
00:02:22,160 --> 00:02:24,760
into the US of course, Canada, 
France, Germany. 

47
00:02:24,760 --> 00:02:27,600
I think those are the countries 
this demand in more places too, 

48
00:02:27,600 --> 00:02:30,520
certainly for this. 
So anyway, so I thought, OK, 

49
00:02:30,520 --> 00:02:31,920
we'll be bold. 
We'll go to San Francisco and 

50
00:02:31,920 --> 00:02:35,120
sponsor hackathon there. 
And I was, you know, I was 

51
00:02:35,120 --> 00:02:38,120
following AI, following agentic 
AI, listening to lots of 

52
00:02:38,120 --> 00:02:41,520
podcasts and trying things out. 
But I did what MCP was at all. 

53
00:02:41,520 --> 00:02:43,680
I just kept hearing about it. 
I thought, what are we going to 

54
00:02:43,680 --> 00:02:47,320
do to go to San Francisco all 
these cool, like I hipsters, you

55
00:02:47,320 --> 00:02:50,680
know, with all their cool AI 
companies based on the Area 75 

56
00:02:50,680 --> 00:02:52,200
hackathon. 
So exactly. 

57
00:02:52,600 --> 00:02:54,120
Yeah. 
And like, I noticed you bell at 

58
00:02:54,120 --> 00:02:56,920
all we're going to do. 
But, you know, I'm pretty brave 

59
00:02:56,920 --> 00:02:58,920
sometimes. 
So I figured we would just go 

60
00:02:58,920 --> 00:03:01,360
there. 
And then I heard about MCP and I

61
00:03:01,360 --> 00:03:03,120
thought we should build an MCP 
server. 

62
00:03:03,120 --> 00:03:05,320
This is like the latest hotness,
just on the fly. 

63
00:03:05,560 --> 00:03:07,720
Like Yolo MCP. 
That was the idea. 

64
00:03:07,720 --> 00:03:09,600
And then there was an engineer 
who knew about it too and wanted

65
00:03:09,600 --> 00:03:12,320
to try it out out and his 
manager thought was a good idea.

66
00:03:12,320 --> 00:03:15,080
So we just built an MCP server, 
mostly him, his name is Akash 

67
00:03:15,440 --> 00:03:18,880
and he did most of the work and 
we just kind of helped him like 

68
00:03:18,880 --> 00:03:20,720
debug a little bit and then so 
you. 

69
00:03:20,720 --> 00:03:23,000
Did this in work essentially 
like a week or something right? 

70
00:03:23,120 --> 00:03:25,560
It was pretty quick yeah, but 
you can do with MCP, as we've 

71
00:03:25,560 --> 00:03:28,000
learned isn't that hard to 
actually use if you know how to 

72
00:03:28,080 --> 00:03:30,080
use it which we didn't of 
course. 

73
00:03:30,280 --> 00:03:32,240
So there were some confusing 
things that happens and 

74
00:03:32,240 --> 00:03:34,840
confusing bugs but basically 
finished it. 

75
00:03:35,160 --> 00:03:37,640
I went down to this hackathon 
give me a talk of the bidding 

76
00:03:37,640 --> 00:03:40,360
like hey you get MCP server. 
Everyone just figured one of the

77
00:03:40,360 --> 00:03:42,720
many AI companies that existed 
people accepted us as one of 

78
00:03:42,720 --> 00:03:45,520
their own and it was actually 
pretty successful People didn't 

79
00:03:45,520 --> 00:03:47,640
use it much in hackathon 
honestly they kind of used API 

80
00:03:47,640 --> 00:03:50,520
instead but having having it 
meant that we were like a serve 

81
00:03:50,520 --> 00:03:52,560
a player in the scene. 
It was kind of like wearing the 

82
00:03:52,560 --> 00:03:55,880
right shirt or the right classes
we should be knew what was going

83
00:03:55,880 --> 00:03:58,040
on with MCP. 
That's how it all began, just 

84
00:03:58,040 --> 00:04:01,440
kind of as a spontaneous idea to
fit into hackathon. 

85
00:04:01,560 --> 00:04:03,760
And so how did it progress? 
Because it sounded like you 

86
00:04:03,760 --> 00:04:08,320
learned a lot from being 
essentially 10 MCP hackathons in

87
00:04:08,320 --> 00:04:10,760
four countries over a very short
span of time. 

88
00:04:11,000 --> 00:04:13,600
I'm kind of curious A, what you 
learned B what you saw that was 

89
00:04:13,600 --> 00:04:17,160
different in the audiences, the,
or even the tenor of how people 

90
00:04:17,160 --> 00:04:19,560
were approaching MCP, how it was
different, you know, like what, 

91
00:04:19,560 --> 00:04:20,360
what did you learn? 
I. 

92
00:04:20,720 --> 00:04:22,960
Just thought it was really good 
topic and I just got really 

93
00:04:22,960 --> 00:04:25,560
interested in the topic myself. 
I mean, tech these days is 

94
00:04:25,560 --> 00:04:27,480
really fascinating. 
It's changing all the time and 

95
00:04:27,520 --> 00:04:29,960
AI is so interesting. 
I got really deep into actually 

96
00:04:29,960 --> 00:04:31,960
how it worked. 
I thought how could this 

97
00:04:31,960 --> 00:04:33,440
possibly happen? 
Like this is an LM. 

98
00:04:33,440 --> 00:04:35,560
It just like says stuff that 
talks to us. 

99
00:04:35,560 --> 00:04:38,520
We know really what it is. 
It takes in tokens as input, put

100
00:04:38,520 --> 00:04:41,360
that tokens as output. 
How does it like use tools? 

101
00:04:41,640 --> 00:04:43,800
And none of the explanations I 
saw talked about how it actually

102
00:04:43,800 --> 00:04:45,440
worked. 
They're all kind of similar and 

103
00:04:45,440 --> 00:04:48,360
talked about the USBC for AI and
didn't talk about what actually 

104
00:04:48,360 --> 00:04:50,200
happened. 
I did a lot of thinking about 

105
00:04:50,200 --> 00:04:52,120
it, investigating, asking 
friends who are more deeply 

106
00:04:52,120 --> 00:04:54,880
involved than I was, and so I 
really understood it well. 

107
00:04:55,120 --> 00:04:57,400
And then I thought, well, I have
to give a talk in June in New 

108
00:04:57,400 --> 00:04:59,360
York. 
I'll give a workshop at MCP 

109
00:04:59,600 --> 00:05:01,440
because I think it should be a 
thing everyone understands how 

110
00:05:01,440 --> 00:05:02,720
to do. 
So this would be a good reason 

111
00:05:02,720 --> 00:05:05,040
for me to figure out like in a 
couple weeks how to really works

112
00:05:05,280 --> 00:05:07,560
well enough to actually teach. 
People had to use it, which is a

113
00:05:07,560 --> 00:05:09,200
big challenge to actually teach 
the thing you don't really know 

114
00:05:09,200 --> 00:05:11,080
at all. 
But again, you try to go for 

115
00:05:11,080 --> 00:05:13,480
challenges in life and this 
makes us learn things. 

116
00:05:13,880 --> 00:05:16,240
So I realized it wasn't that 
hard to actually make a server. 

117
00:05:16,240 --> 00:05:18,600
Like you could actually make a 
server and seven lines of 

118
00:05:18,600 --> 00:05:21,160
Python, you could explain it 
actually pretty easily. 

119
00:05:21,440 --> 00:05:24,360
I found like the inspirations 
for it, like the tool former 

120
00:05:24,360 --> 00:05:26,960
paper from a couple years ago 
that talked about how to 

121
00:05:26,960 --> 00:05:31,320
actually just train the model 
with the data that wasn't just 

122
00:05:31,320 --> 00:05:34,600
data, but also these faux 
function calls that then could 

123
00:05:34,600 --> 00:05:37,320
be spat out by the model later 
on as actual function calls 

124
00:05:37,320 --> 00:05:40,040
intercepted and then uses actual
function calls. 

125
00:05:40,480 --> 00:05:44,480
So I gave this workshop and it 
was surprisingly successful. 

126
00:05:44,720 --> 00:05:47,400
Like the room was packed. 
People had a lot of questions. 

127
00:05:47,440 --> 00:05:48,960
They actually listened. 
I think for the most part they 

128
00:05:48,960 --> 00:05:50,920
didn't ignore me as much as they
might normally ignore me 

129
00:05:51,400 --> 00:05:53,040
afterwards. 
People were like, the next day I

130
00:05:53,040 --> 00:05:55,000
built an MCP server, I did it, 
here it is. 

131
00:05:55,240 --> 00:05:56,480
So I realized. 
There was demand. 

132
00:05:56,480 --> 00:05:58,120
This is great. 
Then you're like, oh. 

133
00:05:59,680 --> 00:06:01,600
Yeah, people really were 
interested in the topic and it's

134
00:06:01,600 --> 00:06:05,240
a nice entry point to learning 
about how to use AI because I 

135
00:06:05,240 --> 00:06:07,000
realized you could build a 
server pretty quickly. 

136
00:06:07,000 --> 00:06:09,680
I made a sample server that was 
just basically it told jokes. 

137
00:06:10,120 --> 00:06:13,280
So the the the simplest tool 
just returned the same joke 

138
00:06:13,280 --> 00:06:15,400
every time. 
So very easy to understand. 

139
00:06:15,400 --> 00:06:17,680
You know, just return the same 
string every time and the joke 

140
00:06:17,680 --> 00:06:20,880
the API found is very easy to 
use The API cube is necessary. 

141
00:06:20,880 --> 00:06:22,760
You can just call the API as a 
REST API. 

142
00:06:23,000 --> 00:06:26,200
So I made more tools to get 
various kinds of jokes and I 

143
00:06:26,200 --> 00:06:27,920
thought it was pretty easy to 
understand how it worked. 

144
00:06:27,920 --> 00:06:29,840
It didn't look like complicated 
code or anything. 

145
00:06:30,120 --> 00:06:33,000
Once you install a fast MZP or 
your favorite client library, 

146
00:06:33,240 --> 00:06:35,760
you can just kind of do it. 
So I made some sample servers 

147
00:06:35,760 --> 00:06:37,440
like this and I started using 
these and more and more 

148
00:06:37,440 --> 00:06:40,600
workshops because people always 
wanted to hear about MCP. 

149
00:06:41,000 --> 00:06:43,160
It was a good topic for talks. 
Obviously. 

150
00:06:43,160 --> 00:06:46,760
So like over the next 7-8, we 
like what changed, what 

151
00:06:46,760 --> 00:06:50,080
progressed is was there like, 
yeah, sort of a, you know, 

152
00:06:50,280 --> 00:06:52,480
things that at the end that were
very different or. 

153
00:06:52,840 --> 00:06:55,200
I did a bunch of Google Dough 
Fest, actually, because I used 

154
00:06:55,200 --> 00:06:57,400
to work at Google. 
I had friends there, so they had

155
00:06:57,400 --> 00:07:00,880
me come in October, actually. 
My, my, my, my indie rock band 

156
00:07:00,880 --> 00:07:04,120
played, opened the show. 
Then I gave MCP Talk after that,

157
00:07:04,400 --> 00:07:06,080
which is kind of a funny 
combination. 

158
00:07:06,320 --> 00:07:11,120
Possibly what changed there were
mostly it was that there were 

159
00:07:11,120 --> 00:07:13,480
cases of people who were more 
interested or less interested. 

160
00:07:13,720 --> 00:07:16,400
People needed to know about it 
more in some cases, less in some

161
00:07:16,400 --> 00:07:18,520
cases. 
But I think the demand is still 

162
00:07:18,520 --> 00:07:21,600
as tied as it was before. 
I did my last workshop on Pure 

163
00:07:21,600 --> 00:07:26,520
MCP in February in Philadelphia,
and it was for an audience of 

164
00:07:26,520 --> 00:07:28,680
the eye specialist who really 
understood the train models more

165
00:07:28,680 --> 00:07:30,400
than I do. 
And they were so interested in 

166
00:07:30,400 --> 00:07:32,360
the talk and didn't know these 
things about how it worked. 

167
00:07:32,360 --> 00:07:35,760
And I mean, I think the main 
thing is that people don't quite

168
00:07:35,760 --> 00:07:37,440
understand what it is that 
easily. 

169
00:07:37,640 --> 00:07:40,480
People think it's a way of 
mixing AI into your API in some 

170
00:07:40,480 --> 00:07:41,760
way. 
Like API is going to change what

171
00:07:41,760 --> 00:07:44,880
it does and magically it's going
to sort of do new things, which 

172
00:07:44,880 --> 00:07:48,640
of course isn't what it is. 
It's AI using AP is. 

173
00:07:48,640 --> 00:07:51,760
So I think they like the idea 
that you can actually get the AI

174
00:07:51,760 --> 00:07:54,880
to act in a rational way, like 
use one of these tools over 

175
00:07:54,880 --> 00:07:57,600
here, but use API other kinds of
ways. 

176
00:07:57,600 --> 00:08:01,200
Yes, restricted, you know, it's,
it's discreet, it's not as 

177
00:08:01,200 --> 00:08:03,520
stochastic. 
But to your question, what 

178
00:08:03,520 --> 00:08:06,560
actually has changed since then?
Not much actually has changed, 

179
00:08:06,560 --> 00:08:07,800
to be honest. 
That's cool. 

180
00:08:08,240 --> 00:08:11,360
So essentially the the like the 
protocol people are starting to 

181
00:08:11,360 --> 00:08:13,120
get in and there's the questions
are better. 

182
00:08:13,120 --> 00:08:16,080
And the desire to learn is still
as big as it was before. 

183
00:08:16,080 --> 00:08:18,000
The questions are kind of the 
same questions like So what? 

184
00:08:18,000 --> 00:08:20,280
Are like the top five questions 
that people get or the 

185
00:08:20,280 --> 00:08:22,840
misconceptions or the I mean, if
you were going to redesign the 

186
00:08:22,840 --> 00:08:25,640
course from the ground up, what,
what would you do differently? 

187
00:08:25,880 --> 00:08:28,120
I would like to show how it 
worked by actually showing an 

188
00:08:28,120 --> 00:08:29,920
example. 
Actually, it's not that hard to 

189
00:08:29,920 --> 00:08:31,720
do. 
You actually can go to Claude or

190
00:08:31,720 --> 00:08:34,799
your LM and just tell us, OK, 
now you can use tools with this 

191
00:08:34,799 --> 00:08:38,760
protocol over here and put in 
some fake XML or Jason and it 

192
00:08:38,760 --> 00:08:41,320
will say, OK, I'll do that. 
I'll begin spitting out things 

193
00:08:41,320 --> 00:08:44,039
using those tools. 
You could then say, what's 5 * 8

194
00:08:44,039 --> 00:08:46,840
use your new math tool and uses 
the math tool it already knows 

195
00:08:46,840 --> 00:08:48,840
about. 
And instead of playing out 40, 

196
00:08:48,840 --> 00:08:53,800
it says, you know, like angle 
brackets math, 5 * 8 close angle

197
00:08:53,800 --> 00:08:56,200
brackets math. 
So now the LMS are good at 

198
00:08:56,200 --> 00:08:57,840
actually using tools and smart 
about it. 

199
00:08:57,880 --> 00:08:59,920
I think I tried to talk about 
how it actually works. 

200
00:08:59,960 --> 00:09:03,080
You know, it actually generates 
a prompt and the prompt is the 

201
00:09:03,080 --> 00:09:04,320
thing which gets in the system 
prompt. 

202
00:09:04,320 --> 00:09:07,200
And this explains how the tools 
work is why your contacts went 

203
00:09:07,200 --> 00:09:08,920
dull and get you, not because 
you have a lot of stuff in 

204
00:09:08,920 --> 00:09:10,760
there. 
I would explain how it worked 

205
00:09:10,920 --> 00:09:13,560
and what it is. 
People wouldn't be as surprised 

206
00:09:13,560 --> 00:09:17,040
by what it actually is. 
So I still get questions about 

207
00:09:17,040 --> 00:09:20,120
that, like how it actually works
and questions that reflect what 

208
00:09:20,120 --> 00:09:23,440
really is going on. 
But more common questions I get 

209
00:09:23,440 --> 00:09:25,360
are about security. 
I was going to set. 

210
00:09:26,040 --> 00:09:29,320
So what do you tell them? 
I tell them be careful because 

211
00:09:29,320 --> 00:09:31,200
it's kind of like shareware. 
I mean, if you're downloading 

212
00:09:31,200 --> 00:09:34,880
someone else's code to your 
laptop and letting it run, then 

213
00:09:34,880 --> 00:09:36,520
be careful about that kind of 
thing on the design. 

214
00:09:36,680 --> 00:09:40,160
Side what are the because I mean
you know the project has a long 

215
00:09:40,160 --> 00:09:42,880
security best practices doc. 
What do you intend to starve 

216
00:09:42,880 --> 00:09:43,960
with? 
And helping people. 

217
00:09:43,960 --> 00:09:46,400
There's a lot of questions you 
get that I can't answer that 

218
00:09:46,400 --> 00:09:48,800
well because they're about are 
there ways to actually like 

219
00:09:48,800 --> 00:09:51,800
inject prompts back into the LLM
or you know, other things. 

220
00:09:52,160 --> 00:09:54,760
But I think part of it is just 
that if you're using an MCP to 

221
00:09:54,760 --> 00:09:57,960
use your API, it's good to not 
expose too much of it and it's 

222
00:09:57,960 --> 00:10:00,280
good to tell us if the API be 
very secure in the first place. 

223
00:10:00,720 --> 00:10:02,120
And then it comes down to API 
security. 

224
00:10:02,120 --> 00:10:05,480
You can use the same kinds of 
things, I think, to secure API 

225
00:10:05,480 --> 00:10:08,480
and secure API against LMS using
your API. 

226
00:10:09,000 --> 00:10:10,800
And then, yeah, look for those 
best practices and 

227
00:10:10,800 --> 00:10:13,480
authorization. 
Make sure that people can't just

228
00:10:13,480 --> 00:10:17,400
like easy spin up gigantic 
queries that will, you know, 

229
00:10:17,400 --> 00:10:20,240
slow down your your servers and 
things like that. 

230
00:10:20,880 --> 00:10:23,040
Honestly, I don't have the best 
advice for security. 

231
00:10:23,040 --> 00:10:24,960
I think it's still a thing 
people have to deal with 

232
00:10:24,960 --> 00:10:26,680
themselves and looked into 
themselves. 

233
00:10:27,120 --> 00:10:31,080
My best advice is still just 
make sure your stuff is secure, 

234
00:10:31,080 --> 00:10:33,520
you know, be careful about it. 
Make sure you do the usual 

235
00:10:33,520 --> 00:10:36,240
things, sanitizing, input, all 
those kinds of things usually 

236
00:10:36,240 --> 00:10:38,120
would do. 
Look out for authorization 

237
00:10:38,120 --> 00:10:41,880
problems, those kinds of things.
And so as you kind of traveled 

238
00:10:41,880 --> 00:10:45,920
from one of these to the next, 
did you see differences in the 

239
00:10:45,920 --> 00:10:49,200
way people thought about it, you
know, like the AI experts versus

240
00:10:49,200 --> 00:10:51,760
the noobs or people that are 
sort of full stack guys? 

241
00:10:51,760 --> 00:10:54,720
Or, you know, what were the 
difference in the questions you 

242
00:10:54,720 --> 00:10:55,960
got from the different 
audiences? 

243
00:10:56,240 --> 00:10:57,760
I couldn't always tell, 
honestly. 

244
00:10:57,760 --> 00:10:59,520
That's a good question. 
I should be able to tell, but 

245
00:10:59,520 --> 00:11:02,240
it's just different places and 
different kinds of questions and

246
00:11:02,400 --> 00:11:04,040
different kinds of levels of 
interest. 

247
00:11:04,200 --> 00:11:07,080
I gave one talk actually at a it
was actually at a Business 

248
00:11:07,080 --> 00:11:10,080
School in Canada. 
It was another Google Dev Fest, 

249
00:11:10,080 --> 00:11:11,520
but the students weren't that 
interested. 

250
00:11:11,760 --> 00:11:13,600
The professionals were more 
interested. 

251
00:11:14,160 --> 00:11:17,200
This talk I gave, I think at all
was API days in Paris. 

252
00:11:17,560 --> 00:11:20,560
The room was really full. 
People were outside the room and

253
00:11:20,560 --> 00:11:22,120
they're really just curious 
about what this thing was, how 

254
00:11:22,120 --> 00:11:24,480
they could use it themselves. 
I think we've estimate how 

255
00:11:24,480 --> 00:11:27,360
people actually might know 
things because people that are 

256
00:11:27,360 --> 00:11:29,520
really involved in a technology 
really don't think everyone 

257
00:11:29,520 --> 00:11:31,640
knows what they're doing. 
But everyone else out there is 

258
00:11:31,640 --> 00:11:34,160
just kind of like, I saw this 
thing, I saw an article about 

259
00:11:34,160 --> 00:11:36,480
this thing somewhere, but what 
is this thing? 

260
00:11:36,480 --> 00:11:38,760
I don't quite know what it is. 
Unless you actually use it, you 

261
00:11:38,760 --> 00:11:41,840
don't really get what it is. 
So if you were going to try to 

262
00:11:41,840 --> 00:11:48,120
summarize like, you know, the 
the quick start guide in at all 

263
00:11:48,200 --> 00:11:50,800
5 minutes or whatever, what 
what, what are the principles 

264
00:11:50,800 --> 00:11:53,320
you'd go over? 
I would just show how to build a

265
00:11:53,320 --> 00:11:55,880
really simple MCP server. 
And yeah, I talked about the the

266
00:11:55,880 --> 00:11:58,320
principles first of like 
basically what it is. 

267
00:11:58,320 --> 00:12:02,920
It's a way you can get an LM to 
actually make these faux 

268
00:12:02,920 --> 00:12:05,160
function calls. 
And you then control the 

269
00:12:05,160 --> 00:12:07,400
function calls yourself. 
It calls your server. 

270
00:12:07,600 --> 00:12:08,960
Your server is working over 
here. 

271
00:12:08,960 --> 00:12:11,760
It's just basically functions. 
It can call your API or not, but

272
00:12:11,760 --> 00:12:13,960
it's all it is, is basically you
write functions of them. 

273
00:12:13,960 --> 00:12:18,040
MCP server can go ahead and call
if it wants to and then show how

274
00:12:18,040 --> 00:12:20,640
to build some simple ones. 
You know, simple simple things. 

275
00:12:21,000 --> 00:12:23,440
Once you get simple ideas down, 
the risk can follow from there. 

276
00:12:23,760 --> 00:12:26,440
But I found harder those MCP 
apps and other kinds of new 

277
00:12:26,440 --> 00:12:28,840
things, they don't always work 
for me quite yet. 

278
00:12:29,000 --> 00:12:31,720
And the harder did you explain 
like the protocol for MCP apps 

279
00:12:31,720 --> 00:12:34,280
is more complicated, MCP UI, 
this sort of thing. 

280
00:12:34,280 --> 00:12:36,520
Yeah, I have not had a simple 
way to actually explain how that

281
00:12:36,520 --> 00:12:38,080
works or to even show how that 
works. 

282
00:12:38,080 --> 00:12:41,840
So were you involved with the 
design and construction at DPL 

283
00:12:41,840 --> 00:12:43,800
of the MCP server? 
I mean, did you kind of watch 

284
00:12:43,800 --> 00:12:44,640
the process? 
Yeah. 

285
00:12:44,640 --> 00:12:46,120
I kind of maintained myself now,
so. 

286
00:12:46,440 --> 00:12:49,360
So, So what, what were the 
conventions around that? 

287
00:12:49,360 --> 00:12:51,560
Because I mean, it's kind of 
interesting that I mean, the 

288
00:12:51,560 --> 00:12:55,720
idea of a language server over 
MCP, yeah, you know, like sort 

289
00:12:55,720 --> 00:12:59,440
of why or what's different or I 
mean, how do you design it, you 

290
00:12:59,440 --> 00:13:01,520
know, so it's different than a 
straight up API or something 

291
00:13:01,520 --> 00:13:03,160
like that. 
Yeah, it's not that different. 

292
00:13:03,160 --> 00:13:05,920
I mean, it's mostly about trying
to find the most commonly used 

293
00:13:05,920 --> 00:13:09,000
functions of the API and make 
them really easy to use, like a 

294
00:13:09,400 --> 00:13:12,360
translation texts, you know, 
sending texts, getting text back

295
00:13:13,040 --> 00:13:16,280
and just trying to make sure you
can look for errors like our 

296
00:13:16,280 --> 00:13:19,320
language codes used correctly. 
Did you require a country code 

297
00:13:19,320 --> 00:13:22,080
over here or not? 
And give helpful information to 

298
00:13:22,160 --> 00:13:23,720
the LM so it knows what to 
actually do. 

299
00:13:23,720 --> 00:13:25,680
And when I first did this, I 
remember it would have failed 

300
00:13:25,680 --> 00:13:27,040
because we use language codes 
incorrectly. 

301
00:13:27,040 --> 00:13:29,440
It would use lowercase and 
uppercase sometimes and things 

302
00:13:29,440 --> 00:13:32,120
like that. 
So it's not that forgiving yet, 

303
00:13:32,560 --> 00:13:34,880
but now it's much more forgiving
because one, we made it more 

304
00:13:34,880 --> 00:13:37,280
forgiving by giving more 
information in the annotations, 

305
00:13:37,480 --> 00:13:40,120
annotations and so on. 
And two, LM service has gotten 

306
00:13:40,120 --> 00:13:42,800
smarter of what they try. 
So when you started, what were 

307
00:13:42,800 --> 00:13:45,000
what were you working on? 
I mean what was the LLM in the 

308
00:13:45,000 --> 00:13:47,720
back what you were piping? 
Into I usually use cloud because

309
00:13:47,720 --> 00:13:50,560
cloud desktop was the most fun 
thing to demo in and it's kind 

310
00:13:50,560 --> 00:13:52,560
of make make conversation with 
you and it would try things out 

311
00:13:52,560 --> 00:13:54,960
and try to be helpful. 
But it's gotten much better at 

312
00:13:54,960 --> 00:13:56,840
this stuff. 
Like last time I gave him this, 

313
00:13:56,840 --> 00:13:59,840
I, I did the talk at Drupacon 
last week and I wrote an MCP 

314
00:13:59,840 --> 00:14:04,000
server that would look for 
translations existing in your 

315
00:14:04,000 --> 00:14:05,880
Drupal site. 
If there weren't any, it was 

316
00:14:05,880 --> 00:14:08,040
just making new ones for you and
send those over for you. 

317
00:14:08,600 --> 00:14:10,440
And it began to do all this 
stuff by itself. 

318
00:14:10,480 --> 00:14:13,840
I gave it a couple of tools and 
it would, it would use DPL's MCP

319
00:14:13,840 --> 00:14:15,080
server. 
It's already using my clawed 

320
00:14:15,080 --> 00:14:17,480
desktop and use my, when I made 
for Drupal. 

321
00:14:17,480 --> 00:14:19,080
And it was just all kinds of 
things. 

322
00:14:19,080 --> 00:14:21,080
Do you want me to translate this
into Danish for you? 

323
00:14:21,400 --> 00:14:23,080
And here it is. 
And it also would comment on 

324
00:14:23,080 --> 00:14:25,880
translations because it also 
knows translations itself being,

325
00:14:26,000 --> 00:14:29,080
you know, also clawed behind the
scenes to comments on the 

326
00:14:29,120 --> 00:14:30,840
translations and other kinds of 
ways did. 

327
00:14:30,840 --> 00:14:33,400
You I mean, so essentially you 
had it almost like Co browsing 

328
00:14:33,640 --> 00:14:36,240
and like, Hey, look at this 
Drupal site and you know, and 

329
00:14:36,480 --> 00:14:39,440
support of the CMS and look at 
sort of all the meta text or 

330
00:14:39,440 --> 00:14:42,400
whatever and and tell me what 
you think it I mean, and then it

331
00:14:42,400 --> 00:14:44,080
went on its own. 
It's like, Oh, by the way. 

332
00:14:44,120 --> 00:14:46,640
Yeah, I give access to look at 
the nodes and see what was in 

333
00:14:46,640 --> 00:14:48,680
the nodes, the article nodes, 
other kinds of those that we 

334
00:14:48,680 --> 00:14:50,600
just look at these things and 
start making suggestions. 

335
00:14:50,600 --> 00:14:53,040
But I didn't expect at all. 
I thought require a lot more 

336
00:14:53,040 --> 00:14:55,640
human intervention but really 
tries to be helpful. 

337
00:14:56,120 --> 00:14:59,120
And and that's like something 
that in the past weren't there 

338
00:14:59,120 --> 00:15:01,880
specialized tools that you 
would, you would, you know, use 

339
00:15:01,880 --> 00:15:05,800
for translation versus, you 
know, sort of generic, you know,

340
00:15:05,800 --> 00:15:07,920
LLM. 
That was the idea exactly 

341
00:15:07,920 --> 00:15:09,640
because people use these things 
called cat tools. 

342
00:15:09,880 --> 00:15:12,000
Whether these large things are 
often very old looking 

343
00:15:12,000 --> 00:15:15,400
interfaces that are kind of like
a scary looking version of Excel

344
00:15:15,680 --> 00:15:17,960
where the send off stuff from 
machine translation has a first 

345
00:15:17,960 --> 00:15:20,480
draft, Then the translator will 
then modify them as they want to

346
00:15:20,480 --> 00:15:23,160
and it can go and find with 
their text you've translated 

347
00:15:23,160 --> 00:15:26,120
before is present in new text 
and they can try to suggest more

348
00:15:26,120 --> 00:15:28,120
things for you. 
So I thought if you're actually 

349
00:15:28,120 --> 00:15:30,240
a translator working for a 
Drupal site, you're not going to

350
00:15:30,240 --> 00:15:32,760
want to go and probably walk in 
and like start typing things as 

351
00:15:32,760 --> 00:15:34,600
a Drupal because it could be a 
little confusing. 

352
00:15:34,600 --> 00:15:37,320
Maybe you're going to want to 
use your favorite LM and just 

353
00:15:37,320 --> 00:15:39,760
talk to it, you know, use claw 
or ChatGPT or something. 

354
00:15:40,240 --> 00:15:42,160
And it was surprisingly 
effective at that. 

355
00:15:42,680 --> 00:15:44,400
Like it was really pretty 
friendly. 

356
00:15:44,720 --> 00:15:47,400
So you almost discovered a new 
product category for deep L 

357
00:15:47,680 --> 00:15:49,560
without realizing what you're. 
Doing yeah, maybe we'll actually

358
00:15:49,560 --> 00:15:51,000
make this into a product people 
could use. 

359
00:15:51,000 --> 00:15:53,600
I mean, I think in these MCP 
apps you could actually. 

360
00:15:53,720 --> 00:15:55,960
What it can't do now that what 
server I made like in a couple 

361
00:15:55,960 --> 00:16:00,080
of days is you, you can't really
modify things yourself that 

362
00:16:00,120 --> 00:16:01,560
easily. 
You can, but it's not. 

363
00:16:01,560 --> 00:16:04,320
It's mostly read only. 
It's like, you know, it'll say, 

364
00:16:04,400 --> 00:16:05,760
OK, here's the translation for 
you. 

365
00:16:05,760 --> 00:16:08,080
Do you want to use this or not? 
You can type a new translation, 

366
00:16:08,080 --> 00:16:09,880
you can copy and paste it and 
change things. 

367
00:16:10,320 --> 00:16:11,720
It should be more like an 
interface, more like a cat to 

368
00:16:11,720 --> 00:16:14,480
where you have little boxes and 
here's the translations. 

369
00:16:14,600 --> 00:16:16,760
Do you like these things or not?
Feel free to modify these 

370
00:16:16,760 --> 00:16:18,920
things. 
Send them all off as a batch to 

371
00:16:18,920 --> 00:16:21,400
the Drupal site. 
That'd be a better way to do it.

372
00:16:22,200 --> 00:16:26,520
From my understanding to the 
audience is understanding how is

373
00:16:26,520 --> 00:16:30,320
Deep L's translation layer 
differentiated or or you know 

374
00:16:30,360 --> 00:16:33,640
better than what you might get 
out of the frontier LMS right 

375
00:16:33,640 --> 00:16:35,120
now. 
Yeah, that's an interesting 

376
00:16:35,120 --> 00:16:37,040
question. 
Because Deep L first came out in

377
00:16:37,040 --> 00:16:39,160
2017. 
They had the best translation 

378
00:16:39,160 --> 00:16:41,040
anywhere because they had their 
own models. 

379
00:16:41,040 --> 00:16:43,480
They trained to do these things 
with linguist help, and it's 

380
00:16:43,480 --> 00:16:45,240
better than Google Translate, 
better than anything else. 

381
00:16:45,720 --> 00:16:47,600
LMS came out with a pretty good 
translation, just kind of 

382
00:16:47,600 --> 00:16:50,480
out-of-the-box. 
So Deep L has now trying to stay

383
00:16:50,480 --> 00:16:53,720
ahead by just taking models, 
building its own models, fine 

384
00:16:53,720 --> 00:16:56,600
tuning models, and just doing 
more and more testing with 

385
00:16:56,600 --> 00:16:58,960
linguists. 
More and more fine tuning and 

386
00:16:59,440 --> 00:17:01,760
studies and tests do show that 
Deep L is better for most 

387
00:17:01,760 --> 00:17:05,119
language pairs in most cases, 
often substantially better. 

388
00:17:05,119 --> 00:17:07,800
But it's more of a race now 
because LMS can do pretty good 

389
00:17:07,800 --> 00:17:09,720
translation. 
If you want very good 

390
00:17:09,720 --> 00:17:11,560
translation, you should use 
though, like a company like Deep

391
00:17:11,560 --> 00:17:13,000
L that does it. 
You know, only this kind of 

392
00:17:13,000 --> 00:17:15,040
stuff. 
People often say they prefer it 

393
00:17:15,040 --> 00:17:17,119
and people will use it because 
it's just better. 

394
00:17:17,520 --> 00:17:20,240
You're translating stuff kind of
casually for fun or just a few 

395
00:17:20,240 --> 00:17:22,400
strings on your site. 
You could probably use an ALM 

396
00:17:22,400 --> 00:17:26,440
out there and do it that way. 
So if anybody wants to use the 

397
00:17:26,480 --> 00:17:30,720
Dibel MTP server, they can 
basically integrate it into like

398
00:17:30,760 --> 00:17:33,000
as a front layer in front of 
anything that they're doing. 

399
00:17:33,200 --> 00:17:36,280
Yeah, just stick it in there, 
try it out, and your LM will 

400
00:17:36,280 --> 00:17:39,080
know to use DIBEL usually 
instead of its own translation 

401
00:17:39,080 --> 00:17:41,200
these days, because it will know
there's a tool for that. 

402
00:17:41,720 --> 00:17:43,680
They're kind of biased towards 
using tools as opposed to not 

403
00:17:43,680 --> 00:17:45,760
using tools. 
Yeah, and just gives you a 

404
00:17:45,760 --> 00:17:48,280
higher quality of translation. 
If you have things where your 

405
00:17:48,280 --> 00:17:50,800
translation quality really 
matters to you, You know it's 

406
00:17:50,800 --> 00:17:53,600
pretty cheap actually to DPL. 
So were people using, were 

407
00:17:53,600 --> 00:17:57,200
people actually building things,
MCP apps or whatever with the 

408
00:17:57,200 --> 00:18:00,120
DPL MCP during this tour? 
I mean, they just start to see 

409
00:18:00,120 --> 00:18:02,000
people try to do translation 
stuff. 

410
00:18:02,120 --> 00:18:03,000
I. 
Can't tell if they're 

411
00:18:03,000 --> 00:18:05,480
coordinated or not. 
We've seen growth quite a bit in

412
00:18:05,480 --> 00:18:08,280
our DPL and CV server usage is 
going growing. 

413
00:18:08,320 --> 00:18:10,440
It's still very small compared 
to the actual API, which is very

414
00:18:10,440 --> 00:18:12,760
large, of course, but is growing
very fast. 

415
00:18:13,320 --> 00:18:15,040
Even folks in that as you were 
building things with it and 

416
00:18:15,040 --> 00:18:16,520
they're making various things 
with it. 

417
00:18:16,920 --> 00:18:19,000
I think they just find it more 
convenient than using the API 

418
00:18:19,000 --> 00:18:22,000
sometimes because our tools are 
more carefully defined. 

419
00:18:22,000 --> 00:18:24,480
You know, like here's this 
translation text, here's 

420
00:18:24,480 --> 00:18:27,320
translating a document. 
And there are more more 

421
00:18:27,320 --> 00:18:29,560
guardrails around these things 
so that I think it's just easier

422
00:18:29,560 --> 00:18:31,120
to use. 
So like essentially the 

423
00:18:31,120 --> 00:18:34,000
flexibility of MCP was something
that as you were traveling 

424
00:18:34,000 --> 00:18:37,360
around people were it's just 
much easier for or people that 

425
00:18:37,360 --> 00:18:39,320
don't want to have to think in 
terms of the structures of an 

426
00:18:39,320 --> 00:18:41,840
API or. 
API is the reverse because API 

427
00:18:41,840 --> 00:18:43,520
may be a whole new thing, like 
what is this API? 

428
00:18:43,520 --> 00:18:45,680
How do I use this thing? 
MCP? 

429
00:18:45,680 --> 00:18:46,840
I think it's just sort of 
simpler. 

430
00:18:46,840 --> 00:18:49,280
Here's the tool you know. 
You give this tool these two or 

431
00:18:49,280 --> 00:18:51,760
three parameters and then it 
gives you the results you're 

432
00:18:51,760 --> 00:18:53,120
looking for. 
Very cool. 

433
00:18:54,200 --> 00:18:58,120
So when's the next tour? 
That's a good question. 

434
00:18:58,360 --> 00:19:00,880
The last talk was just last 
week, but actually trying to 

435
00:19:00,880 --> 00:19:03,720
focus on making some scale 
content like writing more of the

436
00:19:03,720 --> 00:19:06,640
traditional devrel things like 
more documentation, make some 

437
00:19:06,640 --> 00:19:08,720
videos, things like that. 
So I'm taking a little break 

438
00:19:08,720 --> 00:19:12,240
from traveling out to do those 
things and I want to go back out

439
00:19:12,240 --> 00:19:15,080
there and talk again with you. 
More magic AI and more of those 

440
00:19:15,080 --> 00:19:17,400
sorts of topics that are also of
course hot topics, you know, 

441
00:19:17,400 --> 00:19:19,760
building agents, which is super 
hot as you know. 

442
00:19:20,200 --> 00:19:23,280
But MCP, I mean, people still 
don't know what it is in most 

443
00:19:23,280 --> 00:19:25,520
cases. 
Outside of in New York, San 

444
00:19:25,520 --> 00:19:27,360
Francisco. 
Or people that are just 

445
00:19:27,360 --> 00:19:29,400
following these things and who 
are working with it and try it 

446
00:19:29,440 --> 00:19:31,280
themselves. 
But people are still like, what 

447
00:19:31,280 --> 00:19:32,960
is this thing outside the demand
bubble? 

448
00:19:32,960 --> 00:19:35,240
We always forget the demand is 
still there for just, you know, 

449
00:19:35,240 --> 00:19:37,640
introductory content. 
It's always there for 

450
00:19:37,640 --> 00:19:39,880
everything. 
So the next time you do a tour, 

451
00:19:39,880 --> 00:19:40,960
what? 
What are you going to do 

452
00:19:40,960 --> 00:19:43,240
differently when you design your
MCP curriculum? 

453
00:19:43,720 --> 00:19:46,120
I'm going to try to make MCP app
this time to see if I can get 

454
00:19:46,120 --> 00:19:49,120
all the way to making apps and 
using the new fancier features 

455
00:19:49,120 --> 00:19:51,520
and seeing how that goes. 
It's a good question. 

456
00:19:51,520 --> 00:19:55,280
I feel like what I had did 
before, how can I improve it? 

457
00:19:55,720 --> 00:19:57,680
I feel like actually if I could 
get people to do more coding 

458
00:19:57,680 --> 00:20:00,120
themselves, that'd be nice, 
because when they learn things 

459
00:20:00,120 --> 00:20:02,560
for real, I'd have workshops. 
People often would come in 

460
00:20:02,560 --> 00:20:05,040
without a computer and wanted to
listen to a talk. 

461
00:20:05,120 --> 00:20:06,400
Right. 
So they didn't do hands on as 

462
00:20:06,400 --> 00:20:07,760
much actually. 
I first did. 

463
00:20:07,760 --> 00:20:09,400
I thought I'd do a workshop and 
people just came and they were 

464
00:20:09,400 --> 00:20:11,480
looking at me like, OK, talk to 
us. 

465
00:20:11,600 --> 00:20:14,120
They made it into a talk 
spontaneously and they did it 

466
00:20:14,120 --> 00:20:17,760
later in some cases, try it 
themselves. 

467
00:20:17,760 --> 00:20:19,640
I think it's probably better. 
So I think if I was going to do 

468
00:20:19,640 --> 00:20:22,320
it again, if I do do it again, 
I'll try to get folks to 

469
00:20:22,320 --> 00:20:25,160
actually really write some code 
and know that you can use 

470
00:20:25,160 --> 00:20:27,120
Python, you can use TypeScript, 
you can use your favorite 

471
00:20:27,120 --> 00:20:29,840
language. 
You can do it pretty quickly. 

472
00:20:30,080 --> 00:20:32,040
As long as you've got something 
installed on your computer, 

473
00:20:32,040 --> 00:20:34,880
which is an LLM, which is an AI 
client, which maybe you build 

474
00:20:34,880 --> 00:20:36,840
people don't haven't got yet. 
If they have an AI client, they 

475
00:20:36,840 --> 00:20:39,040
can do it. 
You can install your library, 

476
00:20:39,320 --> 00:20:42,200
you can just go and do it. 
And it feels, it feels like 

477
00:20:42,200 --> 00:20:45,360
magic actually, an LMS are 
actually like using your tool 

478
00:20:45,640 --> 00:20:47,880
and you see it doing that and 
then responding with your 

479
00:20:47,880 --> 00:20:50,560
information you've given it is 
really kind of cool. 

480
00:20:51,080 --> 00:20:54,200
I, I mean, I, I personally, I'm 
not a coder. 

481
00:20:54,200 --> 00:20:55,960
I've been, well, I'm a dangerous
coder, right? 

482
00:20:56,000 --> 00:20:57,440
Like I could code stuff you 
would not. 

483
00:20:57,600 --> 00:20:59,280
If one's a coder. 
You would not want me to put it 

484
00:20:59,280 --> 00:21:01,040
in prod, but yeah, it's just 
blowing me away. 

485
00:21:01,040 --> 00:21:03,800
How you know, you can 
essentially go into Claude code,

486
00:21:03,880 --> 00:21:06,800
give give it a spec. 
It'll recommend the tool. 

487
00:21:06,800 --> 00:21:09,240
Or if there's MCP servers out 
there that'll make sense, it'll 

488
00:21:09,240 --> 00:21:11,280
recommend them. 
You know, it's a, it's, it's, 

489
00:21:11,280 --> 00:21:13,640
it's really, really different. 
Yeah, watching like you get like

490
00:21:13,640 --> 00:21:16,720
Claude or Taji and they actually
use your tools in the call to 

491
00:21:16,720 --> 00:21:19,480
three different tools. 
Like I could do a thing, I could

492
00:21:19,480 --> 00:21:21,760
give a demo, which is like, OK, 
I'm going to have it use the 

493
00:21:22,040 --> 00:21:24,280
translation tool now. 
I'm going to ask it to translate

494
00:21:24,280 --> 00:21:27,320
this thing, like find a joke and
then translate the joke. 

495
00:21:27,640 --> 00:21:29,840
We'll call 2 tools. 
People are on the edge of their 

496
00:21:29,840 --> 00:21:31,200
seats. 
They're going to have actually 

497
00:21:31,200 --> 00:21:33,880
like tell, tell a joke, 
translate it into French and put

498
00:21:33,880 --> 00:21:37,160
it into a spreadsheet. 
Three tools he watches Claude 

499
00:21:37,160 --> 00:21:38,600
spins. 
That actually usually does it 

500
00:21:38,600 --> 00:21:40,720
right. 
So it's just, it's exciting, 

501
00:21:40,720 --> 00:21:43,000
it's oddly suspenseful and 
gratifying because. 

502
00:21:43,200 --> 00:21:45,360
Usually it works. 
Thank you very much for coming 

503
00:21:45,360 --> 00:21:46,240
out on the. 
Podcast. 

504
00:21:46,440 --> 00:21:48,640
Thank you. 
All the audience one last time, 

505
00:21:48,960 --> 00:21:51,440
you know, like who you are and 
what you do again in case you're

506
00:21:51,440 --> 00:21:54,040
catching a tail end. 
In case you forgot or missed the

507
00:21:54,040 --> 00:21:56,160
beginning because whatever 
reason you're doing something 

508
00:21:56,160 --> 00:21:59,360
else, I'm Ben Morris. 
I'm from DPL Developer Advocate 

509
00:21:59,680 --> 00:22:02,360
and we're talking about MCP 
servers, how they work and 

510
00:22:02,360 --> 00:22:05,760
Deepel is a language translation
company that uses AI makes on 

511
00:22:05,760 --> 00:22:09,200
models to make language 
translation as good as possible.

512
00:22:09,600 --> 00:22:11,120
Thanks for coming. 
Thanks for having. 

513
00:22:11,120 --> 00:22:13,120
Me enjoy MCP dev solar. 
I will, thanks.

