WEBVTT 00:00.120 --> 00:03.440 Well, we are now on the final home stretch. 00:03.440 --> 00:07.960 This is the penultimate episode, the penultimate lesson. 00:08.280 --> 00:15.840 It is week six, day four, the beginning, the beginning of our capstone project. 00:15.840 --> 00:18.480 Although really we've been doing the capstone project all week. 00:18.480 --> 00:26.320 But this is the conclusion of the project, a a project which is going to bring together OpenAI agents 00:26.320 --> 00:36.200 SDK and MCP into a commercial project that I'm calling autonomous Alpha agents, because this is a project 00:36.200 --> 00:40.000 applied to financial markets, as you have no doubt guessed. 00:40.200 --> 00:45.360 And the idea is that in this project, we're going to do something that is commercial. 00:45.360 --> 00:49.320 It is first and foremost going to be something that we can apply in business. 00:49.640 --> 00:54.920 And most of our projects so far have been somewhat generic in that there have been things like a sidekick, 00:55.240 --> 00:57.840 uh, and something that builds code. 00:57.920 --> 01:03.920 And I wanted to do something which could apply to a business vertical and use some kind of a setup that 01:03.920 --> 01:07.120 you could imagine applying to many other business verticals. 01:07.240 --> 01:14.040 And so we're applying it to financial markets, and we'll be trying to build something that is exciting 01:14.040 --> 01:18.800 from a commercial business point of view, as well as from a technical agents point of view. 01:19.200 --> 01:25.080 There's going to be five different MCP servers that we'll be using, uh, with tools and resources from 01:25.080 --> 01:25.440 them. 01:25.640 --> 01:28.120 There's going to be interactions between agents. 01:28.120 --> 01:31.400 Of course, it's going to be fully autonomous. 01:31.400 --> 01:33.000 You wouldn't have it any other way. 01:33.360 --> 01:38.760 And I need to tell you, do not under any condition, under any circumstances, use this for your own 01:38.760 --> 01:40.440 trading decisions. 01:40.440 --> 01:47.040 This is educational and for learning and applying agents to understanding how to use them for business, 01:47.040 --> 01:50.040 but not actually to use them for trading decisions. 01:50.400 --> 01:55.880 And now for our capstone project we go to the week six folder. 01:55.880 --> 01:58.570 We go to lab number four for. 01:58.970 --> 02:04.490 And welcome to the lab introducing you to our project Autonomous Alpha agents. 02:04.730 --> 02:06.450 Our equity simulator. 02:06.730 --> 02:09.290 And let me tell you all about what we're going to do. 02:09.330 --> 02:15.370 We're going to build an equity trading simulation with three traders and a researcher agent, powered 02:15.370 --> 02:19.850 by an army of MCP servers with tools and resources. 02:19.970 --> 02:21.090 There's five of them. 02:21.330 --> 02:28.890 The homemade accounts, MCP server that we built based on the work of our own engineering team of agents 02:28.890 --> 02:29.810 from week three. 02:30.290 --> 02:31.010 Fetch. 02:31.450 --> 02:35.690 We're using to get web pages via a headless browser that will be running locally. 02:35.890 --> 02:37.250 Memory will use. 02:37.250 --> 02:42.570 In case our agents or traders wish to save their thoughts, they'll be able to share them too. 02:42.610 --> 02:47.970 So this will allow our trading floor to have a kind of shared memory state as they learn about markets, 02:48.810 --> 02:55.450 search via the brave API and then financial data, either the free version or the paid version. 02:55.450 --> 02:58.770 If you wish to Were entertain yourself as I have done. 02:58.770 --> 02:59.890 It's great fun. 03:00.130 --> 03:05.650 I say it's worth 20 bucks, but I realize that everyone's appetite for this will be different. 03:05.930 --> 03:15.490 Um, so look, the goal of today's part of the capstone project is to build a Python module, and that 03:15.490 --> 03:22.050 Python module will be called traders and will contain a trader class, which will be an independent 03:22.050 --> 03:23.370 autonomous agent. 03:23.970 --> 03:29.730 But the way I like to do this is I like to start in a lab in a notebook, because you can iterate over 03:29.730 --> 03:32.610 things and explore and work on it there. 03:32.770 --> 03:38.530 And then over time, we will migrate to a Python module when we're ready to do so. 03:38.690 --> 03:40.970 That is the that's how I like to work. 03:40.970 --> 03:44.090 And so we'll start in the lab and then move to the module in a bit. 03:44.370 --> 03:50.890 But of course I have to mention one more time yet again this warning that you should not use this for 03:50.890 --> 03:53.970 actual trading decisions, just in case I haven't said that enough. 03:54.010 --> 03:55.450 If you're fed up of me saying it. 03:55.810 --> 03:59.890 All right, so there's a bunch of imports to get us started. 03:59.970 --> 04:01.210 You'd expect nothing less. 04:01.410 --> 04:03.010 And a load dot env. 04:03.250 --> 04:09.890 And now the first stage is to gather all of the parameters for our MCP servers. 04:09.890 --> 04:16.130 So we've defined each of our MCP servers as we've gone with one of these, these dictionaries that store 04:16.130 --> 04:18.930 the parameters for how you launch that server. 04:18.930 --> 04:21.050 And now we're going to gather them all together. 04:21.050 --> 04:26.850 So we're starting with a group called the Trader Server Parameters, which is going to have our account 04:26.850 --> 04:34.810 server where we look up account details and then the server memory where it can maintain its memory. 04:35.170 --> 04:40.770 Uh, and we put that into we also want to have a trader tools that we're going to have as an empty list 04:40.770 --> 04:41.530 to start with. 04:41.890 --> 04:43.650 And then here we've got we've got a choice. 04:43.650 --> 04:48.810 It depends on whether you are using the paid financial services approach. 04:48.970 --> 04:56.090 If so, then we're going to add another MCP server which is the paid financial services with the directory 04:56.130 --> 04:58.690 set to this directory here that you should. 04:58.730 --> 05:02.930 Overwrite with the directory that you're using if you're using the paid. 05:02.970 --> 05:05.210 Financial services API. 05:05.410 --> 05:08.730 If you're using the free one, then we need to get the tools. 05:09.130 --> 05:14.610 From the MCP server using open AI and sorry converting it to open AI tools. 05:14.650 --> 05:18.410 Using the client that I wrote, um, and I might replace that. 05:18.410 --> 05:22.010 You may see the code here might be tidied up if I am able to get it to work. 05:22.010 --> 05:26.530 Just using the the open ai mcp client. 05:27.090 --> 05:27.530 Okay. 05:27.570 --> 05:33.090 So with that we will we will collect together these MCP servers and we will collect them together here 05:33.090 --> 05:33.930 the parameters. 05:34.330 --> 05:38.050 And we will collect one a couple more for the researcher. 05:38.050 --> 05:39.610 We're going to collect fetch. 05:39.890 --> 05:42.370 And we're going to collect the brave search. 05:42.490 --> 05:49.170 So fetch you'll remember that's the one that is going to sort of spawn a headless browser on our computer 05:49.170 --> 05:53.930 that we won't see, but it's going to happen behind the scenes and using playwright, and it will use 05:54.060 --> 05:58.740 that to search the web and bring in web pages and and retrieve their contents. 05:59.060 --> 06:01.700 But the more powerful one is brave. 06:01.740 --> 06:07.900 Brave is where it will call out to the brave search engine, which will run a search and bring back 06:07.900 --> 06:10.220 information relevant to that search. 06:10.500 --> 06:16.460 So these are the two tools, just a web query and a web search, or a fetch and a search. 06:16.740 --> 06:18.740 So let's bring them in as well. 06:18.780 --> 06:21.900 And now we have gathered all of our parameters. 06:21.940 --> 06:22.220 Okay. 06:22.260 --> 06:26.340 Next up we want to actually create the MCP server studio. 06:26.380 --> 06:29.940 The standard input output for each one of our params. 06:30.140 --> 06:33.740 So what we're going to do is we're going to have a list of like let's start with the researcher. 06:33.780 --> 06:35.300 So we're going to have a researcher. 06:37.340 --> 06:39.540 I hate it when cursor fills it all in. 06:39.540 --> 06:41.020 For me that's going to take away the fun. 06:41.180 --> 06:43.340 The researcher MCP servers. 06:43.580 --> 06:46.540 The MCP servers for the researcher is going to be. 06:47.100 --> 06:48.500 And it's got it wrong which is nice. 06:48.540 --> 06:58.380 MCP server standard input Output, um, that we're going to want to create given the params. 06:58.540 --> 06:59.700 And it's got it all right. 06:59.700 --> 07:01.980 Now it's given the parameters. 07:01.980 --> 07:04.620 And we're going to set the cache tools list to true. 07:04.660 --> 07:10.420 We do want to cache them so it doesn't have to create them every time for the parameters in this set 07:10.460 --> 07:10.980 here. 07:10.980 --> 07:13.740 And then we just want to repeat this basically a second time. 07:13.740 --> 07:17.420 But this time we want it to be for the trader parameters. 07:18.020 --> 07:27.140 Um, for those we're going to want to come up with a trader MCP server. 07:27.140 --> 07:34.300 So now we're creating a trader MCP server for every set of parameters in the trader MCP server parameters. 07:34.340 --> 07:34.860 You see what I mean? 07:34.860 --> 07:38.460 We're just iterating through these sets of parameters and creating the servers. 07:38.500 --> 07:38.980 All right. 07:39.020 --> 07:41.660 And then we will just say MCP servers. 07:41.700 --> 07:42.980 Is those two added together. 07:42.980 --> 07:45.420 Those two lists together make all of our servers. 07:45.420 --> 07:52.220 So now we have created our MCP server Stdio for all of these.