WEBVTT

1
00:00:03.210 --> 00:00:05.070
Hi, my name is Tobias Boelter.

2
00:00:05.070 --> 00:00:08.070
I'm currently the head
of security at Harvey AI.

3
00:00:08.070 --> 00:00:10.170
In traditional security,
when you're thinking

4
00:00:10.170 --> 00:00:11.910
about securing your product,

5
00:00:11.910 --> 00:00:13.800
securing the infrastructure, et cetera,

6
00:00:13.800 --> 00:00:16.890
you review the code,
you review the system,

7
00:00:16.890 --> 00:00:18.180
you review the architecture,

8
00:00:18.180 --> 00:00:20.610
and then you move it into production.

9
00:00:20.610 --> 00:00:22.830
Now, I think security
teams need to shift right.

10
00:00:22.830 --> 00:00:25.740
And you need to monitor
what the agent is doing

11
00:00:25.740 --> 00:00:27.870
and respond to it at runtime.

12
00:00:27.870 --> 00:00:29.670
And so this is going to require

13
00:00:29.670 --> 00:00:32.670
quite a different
architecture, set of practices,

14
00:00:32.670 --> 00:00:34.440
set of tools for security teams

15
00:00:34.440 --> 00:00:36.840
to be able to work on that new paradigm.

16
00:00:36.840 --> 00:00:38.760
Your AI agents have the same permissions

17
00:00:38.760 --> 00:00:40.140
as your best engineer.

18
00:00:40.140 --> 00:00:42.150
They can read secrets, access production,

19
00:00:42.150 --> 00:00:44.250
and make potentially risky changes.

20
00:00:44.250 --> 00:00:46.920
Today, you have limited
telemetry and guardrails,

21
00:00:46.920 --> 00:00:49.860
and one incident can set
your AI rollout back.

22
00:00:49.860 --> 00:00:52.350
MCPs, as you know, have been adopted

23
00:00:52.350 --> 00:00:56.190
by pretty much all the main
foundation model companies,

24
00:00:56.190 --> 00:00:57.930
and so it's becoming the standard way

25
00:00:57.930 --> 00:01:00.540
to access tools remotely.

26
00:01:00.540 --> 00:01:03.000
For security, it's always
important to look ahead

27
00:01:03.000 --> 00:01:06.270
and be proactive, versus waiting
for an incident to happen.

28
00:01:06.270 --> 00:01:08.010
MCP has been widely adopted.

29
00:01:08.010 --> 00:01:11.280
It's how agents access the data
they need to get work done,

30
00:01:11.280 --> 00:01:14.421
but there are tens of thousands
of MCP servers out there.

31
00:01:14.421 --> 00:01:16.552
MintMCP helps your team figure out

32
00:01:16.552 --> 00:01:19.320
which ones are secure and actually useful.

33
00:01:19.320 --> 00:01:21.480
The team really, really,
really liked the concept

34
00:01:21.480 --> 00:01:24.510
of virtual MCPs because they were able

35
00:01:24.510 --> 00:01:27.210
to abstract the way some of the complexity

36
00:01:27.210 --> 00:01:31.380
of like which MCPs need to be
added with that virtual MCP.

37
00:01:31.380 --> 00:01:34.110
The other thing that they really liked

38
00:01:34.110 --> 00:01:37.950
is the focus on making
sure all the security

39
00:01:37.950 --> 00:01:40.541
and all the auth tokens
and oath flows happen

40
00:01:40.541 --> 00:01:43.050
through this central gateway.

41
00:01:43.050 --> 00:01:45.360
And then lastly, ease
of use of the product

42
00:01:45.360 --> 00:01:48.123
and setting it up was really,
really straightforward.

43
00:01:48.123 --> 00:01:49.890
MintMCP gives you a visibility

44
00:01:49.890 --> 00:01:53.520
and control over every AI
agent in your organization.

45
00:01:53.520 --> 00:01:55.020
In your first week, you will see

46
00:01:55.020 --> 00:01:57.210
every single MCP server installed,

47
00:01:57.210 --> 00:01:58.770
every tool that was called.

48
00:01:58.770 --> 00:02:00.270
Then you can set guardrails

49
00:02:00.270 --> 00:02:02.040
that match your security posture.

50
00:02:02.040 --> 00:02:03.390
For agents to be useful,

51
00:02:03.390 --> 00:02:05.790
they need to know how your business runs.

52
00:02:05.790 --> 00:02:08.760
Custom MCP servers let
you build that logic in,

53
00:02:08.760 --> 00:02:11.040
so that agents understand
the way that you work.

54
00:02:11.040 --> 00:02:14.790
Static security rules cannot
keep up with dynamic agents.

55
00:02:14.790 --> 00:02:17.299
MintMCP watches how
agents actually behave,

56
00:02:17.299 --> 00:02:18.750
learns the patterns,

57
00:02:18.750 --> 00:02:21.000
and adapts your guardrails automatically.

58
00:02:21.000 --> 00:02:23.310
AI agents move faster than humans,

59
00:02:23.310 --> 00:02:24.780
your governance should, too.

60
00:02:24.780 --> 00:02:25.863
That's MintMCP.