WEBVTT

00:00:00.100 --> 00:00:03.820
Now, earlier in this section,
I already pointed you at this

00:00:03.880 --> 00:00:07.740
tools configuration option here,
which allows you to control which

00:00:07.800 --> 00:00:09.880
tools your agents have available.

00:00:10.080 --> 00:00:14.060
Now, there are many built-in tools
that come with Visual Studio Code,

00:00:14.100 --> 00:00:17.220
but there also are some extra tools
that can be added.

00:00:17.280 --> 00:00:21.160
For one,
certain Visual Studio Code extensions add

00:00:21.200 --> 00:00:24.960
Python extension.
It adds some Python-specific tools,

00:00:25.020 --> 00:00:28.340
like a tool to set up a Python virtual
environment.

00:00:28.360 --> 00:00:31.480
And depending on
which extensions you have installed

00:00:31.520 --> 00:00:35.420
you're working,
you will have more tools from these

00:00:35.440 --> 00:00:38.540
But then there also are MCP servers.

00:00:38.580 --> 00:00:42.420
Now,
MCP stands for Model Context Protocol

00:00:42.500 --> 00:00:46.470
open protocol for exposing tools
and resources

00:00:46.540 --> 00:00:50.020
to AI agents so that those agents can use

00:00:50.080 --> 00:00:54.020
them.
So it's a standardized way for adding

00:00:54.080 --> 00:00:57.980
say. It's not GitHub Copilot exclusive,
but GitHub Copilot

00:00:58.060 --> 00:01:01.960
also supports it.
Inside Visual Studio Code,

00:01:02.000 --> 00:01:05.960
into your command panel here,
you can list MCP servers

00:01:06.020 --> 00:01:09.030
you a- got installed
and you can add a new server.

00:01:09.080 --> 00:01:12.900
If you chose Add new server,
you can also browse MCP

00:01:12.980 --> 00:01:16.380
servers and you can do
that from inside Visual Studio Code.

00:01:16.400 --> 00:01:20.380
And you'll find various MCP servers you
can install, and again, these

00:01:20.420 --> 00:01:24.290
are really just packages of tools
and resources that are then

00:01:24.340 --> 00:01:28.330
made available to your agents.
For example, context7 is a pretty

00:01:28.340 --> 00:01:32.200
good one. Context7 is an MCP server
that allows the

00:01:32.260 --> 00:01:35.480
AI to browse the documentation of many
libraries and

00:01:35.540 --> 00:01:39.360
packages, not just limited to Python,
you might be interested

00:01:39.480 --> 00:01:42.820
in. So by installing this MCP server,
you make it

00:01:42.920 --> 00:01:46.580
easier for your AI agents to discover
and use

00:01:46.640 --> 00:01:49.530
documentation pages off third-party
libraries and

00:01:49.620 --> 00:01:53.400
packages.
So that's one server I have installed.

00:01:53.440 --> 00:01:57.400
You could use the Playwright server to
make it easy for your AI

00:01:57.460 --> 00:02:01.200
agents to access your browser
and test a website

00:02:01.340 --> 00:02:03.840
they might be working on, and so on.

00:02:03.860 --> 00:02:07.560
You can browse these servers
and add whatever you find useful.

00:02:07.620 --> 00:02:11.060
You just shouldn't enable too many,
because all

00:02:11.120 --> 00:02:14.960
these instructions about which tools
are available

00:02:15.060 --> 00:02:18.609
are loaded into the context window
and therefore take up tokens.

00:02:18.620 --> 00:02:22.500
So the more MCP servers you enable
and use, the

00:02:22.660 --> 00:02:26.170
more your results may suffer.
That's at least something I

00:02:26.200 --> 00:02:29.980
experienced. So here, I only got two MCP

00:02:30.060 --> 00:02:34.049
servers enabled for my agent, and
if I had more installed but I

00:02:34.080 --> 00:02:37.829
didn't want to use them all,
I could simply disable them so

00:02:37.829 --> 00:02:41.620
not be loaded into context.
All those tool instructions would not be

00:02:41.660 --> 00:02:45.210
loaded into context for the tools exposed
by that server.

00:02:45.240 --> 00:02:48.160
You can also do this on a per-tool basis,
which is pretty

00:02:48.200 --> 00:02:49.820
useful.