Build and Deploy UiPath Agents With Claude Code (Full Transcript)

[00:00:00] Speaker 1: If you've ever been asked to build some kind of automation for your company, route emails, process invoices, classify tickets, connect two systems together, but the last thing you wanted was to stop everything and learn a completely new platform just to ship it, this video is for you. Because what we're gonna look at today is how you can use Cloud Code, the same AI coding agent you might already be using, to build, verify, and deploy professional-grade automations on UiPath without ever needing to become a UiPath expert yourself. Hi, I'm Garrick, and in this video, in partnership with UiPath, we're gonna walk through how to connect Cloud Code to the UiPath platform and use it to build a working AI agent from a single prompt in just a few minutes. We'll cover the one-time setup, the build process, and then run the finished agent so you can see the whole thing working end-to-end. First, a quick bit of context on UiPath in case you haven't worked with it before. UiPath is one of the leading enterprise automation platforms in the world, and it's what large organizations use to automate repetitive business processes, things like processing invoices, routing customer requests, syncing data between systems, or triggering actions based on incoming emails. Traditionally, building on UiPath meant learning their visual workflow editors, their deployment tools, their whole ecosystem, which is great if automation is your core job, but if you're a developer who just needs to ship one automation and then get back to your actual work, that learning curve has always been a real barrier. But with the latest updates from UiPath, that's starting to change. This is called UiPath for Coding Agents. It's a platform-wide capability that lets an AI tool like Cloud Code build and deploy automations directly on UiPath. And I want to be clear about what build on UiPath actually means because it covers a lot of ground. You can build AI agents, automate UIs through RPA, connect to systems via reusable API workflows, test enterprise apps, process documents intelligently, build your own apps, and then orchestrate entire processes that include all of the above, plus any AI agents you've already built elsewhere. We're going to focus on building an agent today, but the approach is the same across all of these. So here's the concept behind all of this, and it's actually pretty elegant. UiPath has created a set of agent skills, specialized instruction files that get installed directly into coding agents such as Cloud Code, and these skills teach Cloud Code everything it needs to know about UiPath patterns, project structures, and best practices. So when you open Cloud Code and give it a prompt to create an automation for you, it's not just generating generic code. It's building a project that follows UiPath-specific structure using UiPath SDK in a format that the UiPath platform knows how to run, evaluate, and deploy. And at any point, you can switch to UiPath Studio, their visual editor, and continue from there. Think of it like giving Cloud Code a specialized manual before it starts building. It knows the rules it's supposed to follow before it touches anything, and the result is that you can describe what you want in plain English, and Cloud Code acts like a professional automation developer that knows all the implementation details, while UiPath handles the infrastructure, the deployment, and the enterprise-grade guardrails. Before we get into the setup, here's what you'll need. A UiPath account. There's a free community plan that includes everything required for this tutorial, and I'll link to that in the description. You'll need Cloud Code, which runs in your terminal. You can use this with a Cloud Pro or Mac subscription or with an Anthropic API account. You'll need Node.js 22 or higher. 24 is recommended, which we'll use to install the UiPath CLI. And you'll need Python 3.11 or higher, along with a fast Python package manager called uv. Now, I'm on a Mac, but I'll call out the differences for Windows throughout, and the steps are essentially the same. There are just a few platform-specific adjustments along the way. Okay, let's start our installations. If you want to follow along with the full documentation as we go, I'll link to it in the description under Resources. I've opened up Terminal here on my Mac. If you're on Windows, open Git Bash instead. PowerShell and Command Prompt won't work with this particular workflow. You'll also need to install Git for Windows first if you haven't done so already, and I'll link to that in the description as well. So the first thing we need is Cloud Code. It's installed through NPM, which comes with Node.js. So if you don't have Node.js yet, grab version 22 or higher at nodejs.org first. Once Node is installed, run this. Now, the G flag means global. It makes the Cloud Command available everywhere on your machine, not just in one folder. And the Allow Scripts flag tells NPM to allow the install script that's included with Cloud Code that detects the operating system and adds the Cloud Command to the path that you want your users to execute. So we'll run that. You may see some notices about a newer version, and those are just harmless so we can ignore those and move on. But once Cloud Code is installed, let's go ahead and log in now so it's ready when we need it. So we'll just run Cloud to launch it. So this is Cloud Code's first run setup screen. Basically, it's asking you to pick a theme. You can select one here. Dark mode is already selected, so I'll just press Enter. And we can choose our login method here. Cloud Account with Subscription, Anthropic Console Account, or a third-party platform. In my case, I'm going to log in with my subscription and it's going to open up my browser so I can authorize that connection. We get a confirmation that it's all set up. It tells me that the login is successful here and I can press Enter to continue. We see a couple of security notes that show us the recommended settings. I'll say, yes, trust this folder. So basically, just follow all the prompts to log in and set up Cloud Code. And once we're logged in, we'll type exit to close Cloud Code and we'll move on to the next installation. Next, we're going to install the UiPath CLI. That's the command line tool that lets us connect to UiPath, install skills, and manage our project. And we'll install it by running this line. So this installer does a few things. It installs the CLI itself and then automatically installs a set of instruction files, the skills, directly into Cloud Code. So from this point on, whenever you open Cloud Code inside a UiPath project, it'll automatically load these skills and know exactly how to build the UiPath automations. It's asking for my password here. Now, the installer also takes care of two other dependencies, .NET SDK and Python. You may see Homebrew kick in and download those. This is a lot of output and it'll ask for your password, like we saw, to install .NET, and that's all expected. Just basically let all of this run. Once it finishes, you should see the summary at the bottom confirming that the CLI, skills, .NET, and Python are all installed. Next, run this to add the coded agent tool. This adds the commands that we'll use later. And there it is. Now, if you saw checking for updates or updating UiPath CLI before the success message, that's normal. It just means the installer updated itself before running the command. Now, again, if you're on Windows, the UiPath CLI install command needs to be run in PowerShell, not the git bash terminal, and the install command is a little different. I'll link to the UiPath documentation in the description where you can find the Windows version. Next, we'll install uv, which is the Python package manager we'll use to set up our project. On Mac, you'll install it with this curl command. And once it finishes, we'll run this. And that makes the uv command available in your current terminal session. Now, Windows users, you can install uv with pip install uv. And finally, let's log into UiPath from the terminal. We'll run this. And that walks you through a browser window where you'll be asked to log into your UiPath account and select your tenant. Now, I've already logged into my account on this browser, so it's telling me that I'm already in and I can go back to terminal and get to work. Okay, now let's create the local project and set everything up right here in the terminal. We'll run this. This command creates a new folder called support-ticket-agent and immediately navigates into it. And then initialize a Python project with this flag, and this will try to get your name and email from your local git configuration as part of generating the project's file. So you can see it's initialized a project called support-ticket-agent. Then we'll add the UiPath langchain package. Langchain is a popular framework for building AI applications, and UiPath's version of it gives our local code access to their SDK and platform services. Okay, so what we're seeing here is a Python version conflict. The project was set up expecting Python 3.9 or higher, but the UiPath langchain package now requires Python 3.11 specifically. So we need to tell our project to use 3.11. So first we'll run uvpython install 3.11 to install it, and then uvpython pin 3.11 to set it as the version for this project. Then we'll open pyproject.toml, which opens it up in our text editor, and we'll change requiresPython 3.9 to 3.11. Save that. And once that's saved, we'll rerun the langchain installation. And now the install has completed successfully. Now this is something you may or may not see depending on your machine's Python setup, but it's a pretty quick fix. Now we're gonna activate the virtual environment. That's the isolated Python environment UV just created for the project. This is also what connects your local project to a project on UiPath's side, so there's nothing to set up manually first. So this line activates the virtual environment we just created, so the rest of our commands can run inside it. This one tells the UiPath CLI where to find Python in our project. And this sets up the UiPath project structure. It creates the config files that connect our local code to UiPath. And then we'll run the install. So you'll see a list of files being created, configuration files that connect your local project to UiPath, including everything needed to publish later. Now that last line, created agents.md files, and Cloud Code Skills, is the UiPath skills being wired into this specific project. So at this point, we have all of our tools installed, we're logged into UiPath, and we have a local project set up and ready to go. Now we're gonna open up Cloud Code and have it build our agent for us. Okay, so this is where everything comes together. We're about to launch Cloud Code and give it a single sentence describing what we want to build. So there's no writing code manually, no looking up UiPath documentation, just a plain English description and Cloud Code does the rest. So we'll start by running Cloud. And because we previously logged in, we're ready to go here. So I'm gonna add a prompt of build a support ticket classifier of build a support ticket classifier agent in support-agent.py. It should take a support ticket as input with a subject and message body and return a category like billing, technical, account, or general, and a priority level, high, medium, or low, based on the content. Use the UiPath coded agent patterns for the project skills. So there's our prompt. We'll run that. So you see this line here near the top of the response, skill, new agent. So the UiPath skill we installed earlier automatically loaded because we're inside a UiPath project. Cloud Code now knows exactly how to structure a UiPath agent without us having to explain it. So Cloud Code writes a complete Python agent, registers it in the project configuration, and then it runs a test on its own. You may see it catch and fix a small issue, which is completely normal, and it's actually a good demonstration of how Cloud Code works. It doesn't just write code and then stop. It verifies and it self-corrects. And the whole thing took just a couple of minutes. Now, before we can test the agent locally, we need to authenticate the Python SDK so it can reach UiPath's LLM service. So we'll run this from inside Cloud Code. That again opens up a browser window, and it'll ask you to log into your UiPath account if you're not already logged in, and select your tenant the same way we did earlier. Once it confirms that authentication, you're ready to go. So let's give it a test ticket. We'll run the command from inside Cloud Code by prefixing it with an exclamation point. If you prefer to run it directly in the terminal, you can exit Cloud first by typing exit and pressing enter, and then run the same command without the exclamation point at the top. So in this example, we're simulating that we received an email that said, charge twice in the subject line, and duplicate charge on my card in the body. So here's something interesting. Notice that Cloud Code actually caught its own mistake. It built the agent using a field called message, but our test used body, so it threw an error. Cloud Code recognized the mismatch, fixed it automatically, and re-ran the test, this time returning category is billing, priority high. And that's exactly the right answer for a duplicate charge ticket. And this kind of self-correction is one of the things that makes Cloud Code generally so easy to use. It is one of the things that makes Cloud Code genuinely useful as a coding assistant. So here it's asking us to confirm the fix. I'll type rename message to body, and let's make that change. And it confirms that it made that change. So a duplicate charge is a payment issue that needs immediate attention. The agent read the subject and body, understood the context, and made a judgment call. Now let's try a different type of ticket to make sure it's distinguishing between categories correctly. This one has a subject of cannot log in, and a body of my password reset is not working. And this categorized that request as account, which is correct, and the priority set to medium. And it even gives us reasoning, which is the issue involves login access and a password reset problem, which affects the customer's ability to access their account, but does not indicate an immediate business critical impact or outage. Which makes sense, a password reset issue is inconvenient, but not urgent, so medium is a reasonable call here. So at this point, we have a working agent that can take any support ticket, figure out what kind of problem it is, and tell you how urgent it is, all in a couple of seconds, with no manual triage required. And again, this was built from a single prompt in just a couple of minutes. Now in a real deployment, you would connect this to your ticketing system, Zendesk, Freshdesk, or whatever you use, so tickets flow in automatically. The agent classifies each one instantly, and you can wire the output to route tickets to the right team, set priority flags, or trigger follow-up workflows in UiPath. So we built an agent here today, but the same workflow applies to everything else on the UiPath platform. If you want to build an RPA workflow, open Cloud Code in a UiPath RPA project and describe what you want. Something like monitor my inbox and route emails to folders based on content. Cloud Code will load the RPA specific skill and generate the right structure. Or if you want an API workflow, same thing. You can say connect this webhook to send a Slack notification when a new ticket comes in. Cloud Code knows the patterns. Now let's get this agent published to UiPath so it's available in the cloud. First, if you're still inside Cloud Code, type exit and press enter, then head over to studioweb at cloud.uipath.com. Here we'll click create new. We'll select agent. We'll select coded agent and click set up coded agent. And that'll bring you to the setup screen. Here, look at the sync from your IDE into studio section and you'll see a line starting with UiPath underscore project underscore ID. And that's followed by your project ID. So we'll copy that entire line. Next, we need to open up the .env file in a text editor. Back in terminal, make sure you're still inside the support ticket agent folder. You should see it in the prompt. If not, you can run CD support ticket agent to get back there like I did. And in the Mac, we'll run this line. On Windows, you'll run notepad.env. And you'll see the file already has some content from the authentication step, which is just fine. Just scroll to the very bottom here and paste in the line we just copied on its own line and then save the file. And then back in terminal, we'll run UiP coded agent push. All right, so this failed because we got this message saying project authors cannot be empty. That means we have to open up the pyproject.toml file again. And above the description, we'll add this line, but with your actual name in it. We'll save the file. We can close it. And we'll run UiP coded agent push one more time. And now if we go back to Studio Web, we can see the entry point. You'll see your email under the last push by field and version 1.00 all here in the right panel. That version number is a confirmation that it landed. So from here, you can schedule the agent, you can trigger it via API, or even drop it into a larger UiPath workflow. So there you have it. If you want to try this for yourself, you can sign up for UiPath's free community plan. And again, that link is in the description. It includes everything we used in this video. UiPath also has a full course on this on UiPath Academy and a GitHub repo with sample projects you can build on. Both of those are linked below as well. And if this was helpful, please hit subscribe and we'll see you in the next video.