- A Google Cloud developer advocate demos how to use Claude Code with Claude models on Google Cloud to build and deploy a feedback application end to end, role-playing five personas across the software development life cycle.
- Setting up Claude on Google Cloud is streamlined via Application Default Credentials (ADC) plus a wizard that detects your project, region, and available models — no API keys to rotate, per-token billing, provisioned throughput, and data staying in your project.
- The demo combines Claude Code features (plan mode, MCP servers, skills, sub agents, security review) with Google Cloud components (Developer Knowledge API, Cloud Run, Firestore, BigQuery, Looker) to ship a live app without prior GCP deployment expertise.
Building with Claude on Google Cloud
- Claude can augment every persona in the SDLC — PM, UI/UX designer, software engineer, security engineer, and data analyst — replacing slow back-and-forth handoffs.
- The simplest way to use Claude models on Google Cloud is Application Default Credentials, which auto-detect your credentials and environment; a setup wizard then detects project, region, and available models to pin.
- Reasons to run Claude on Google Cloud include per-token billing (no message cap), provisioned throughput for production, no API keys to rotate, project-scoped policies, data staying in your project, and multi-region (global/regional) endpoints.
- Plan mode lets Claude think and propose its approach before writing code, giving you a chance to adjust against standards (e.g. design specs pulled in via a Figma MCP server).
- Google Cloud's Developer Knowledge API (with its MCP server) feeds fresh GCP documentation into Claude Code so it can figure out the best architecture, while Google Cloud Skills cover concrete implementation blocks like deploying to Cloud Run or wiring Cloud Run to Firestore.
- Sub agents let you parallelize implementation — the demo spins up separate agents for the API, the ingestion pipeline, and the dashboard, mirroring a team sprint.
- The reference architecture deploys a feedback API on Cloud Run, stores raw responses in Firestore, post-processes data in BigQuery, and reports in Looker.
- A pre-built security review in Claude Code checks for common issues (e.g. OWASP risks, over-privileged service accounts), auto-fixes findings, and then deploys the secured backend.
Application Default Credentials (ADC) — A Google Cloud authentication method that automatically discovers your credentials based on the environment, avoiding manual API key handling.
Provisioned throughput — Reserved model capacity on Google Cloud for enterprise applications that need predictable performance in production.
MCP server — A Model Context Protocol server that exposes tools and data (e.g. fresh documentation) to Claude Code.
Developer Knowledge API — A Google Cloud service that provides up-to-date GCP documentation consumable by Claude Code through an MCP server.
Skills — Pre-built capability blocks that teach Claude how to perform concrete implementation tasks, such as deploying an API on Cloud Run.
Sub agents — Parallel Claude agents that each own a component of the build, enabling concurrent implementation.
Plan mode — A Claude Code mode where the model proposes its plan before writing any code, allowing review and adjustment.
Cloud Run — A serverless Google Cloud service for deploying applications and APIs.
BigQuery — Google Cloud's analytical data warehouse used for post-processing and analyzing collected data.
Hello everyone. Thank you for being here today. I'm pretty excited to have this session with you. I know we are almost at the end of the event, so thank you for bearing with us. To start, let me introduce myself. I'm Ivan Nardini. I'm a developer advocate at Google Cloud working on building content in partnership with Anthropic. To begin this presentation today, I want to ask a very simple question: how many of you in the last week used any AI tool to code or build an application? Okay, the majority. How many of you use the same AI tool to build and deploy applications on Google Cloud? Yeah, just a few. So the goal here today is just to try to make it better. In this demo, what I'm going to show you is how you can use Claude on Google Cloud to build and deploy an application end to end. To do that, I'm going to wear five different hats.
We will start by imagining the use case, the scenario. You are in an enterprise context. In an enterprise context, you probably have a team like the one we're visualizing here that is engaged to build a new feature or a new product. Starting from the left, you probably have a PM. The PM might have an idea of how to improve a particular product or how to implement a feature. Starting from this idea, he shares it with a UI/UX designer who allows him to design and visualize the idea. After the idea starts getting a shape, it is sent to a software engineer who starts developing the core logic behind it in order to ship and make the application accessible to everyone. Before shipping it, in order to be confident in what you are releasing, you probably have to pass a security review. That's why you have a role like the security engineer that allows you to be confident in the release. Finally, once the application gets released, you probably have a data persona like a growth marketer or a data analyst who analyzes the data collected through the app in order to generate insights and produce feedback to the PM so that the product or feature gets improved. These are all the personas that, in a simple way, we can imagine are involved in the software development life cycle. With respect to these personas, Claude — the Anthropic coding agent — can augment all of them, providing several components that we are going to use today.
For the remaining part of this presentation, I'm going to put on the hat of all these personas. I'm going to show you how you can use these Claude Code components together with Claude on Google Cloud in order to build and ship a very simple feedback application that, at the end of the presentation, we are going to use to rate my performance here. But before we start building, you need to set up Claude Code. I'm so proud that we worked together with Anthropic to make this process of setting up Claude Code to use models on Google Cloud very simple and straightforward. You have multiple methods to use Claude models on Google Cloud in Claude Code, but the simplest and fastest one is using Application Default Credentials, which automatically finds your credentials based on the environment that you're going to use. Recently, Claude Code also introduced a wizard that allows you to detect your project and your region where the models are served, check which models are available in your project, and let you pin them in order to start building your application.
At this point you're wondering, "Okay, but what's different about using Claude Code with Claude models? Why use Claude on GCP?" There are many reasons. First of all, you pay for what you use — the usage of Claude models on Google Cloud is per token, so you don't receive a message cap. Also, if you're building an enterprise application that needs to go to production, you can always access what is called provisioned throughput, which reserves some throughput for you. The other important reason is that the setup is straightforward using ADC. You don't have APIs to rotate or environment variables to set. You can access models in your project with your own policies set, and the data stays in your project while you are interacting with Claude Code. Models are served in multiple regions, so you have global endpoints and regional endpoints depending on the availability that you need. As Google Cloud, we have great availability service standards that allow you to use Claude on one of the most performing infrastructures in the market. These are some of the main reasons why you want to consider Claude on Google Cloud, especially in an enterprise context.
Now that you have a few reasons, we are ready to build. As I said, I'll start by wearing the hat of a PM. Imagine you want to improve our services or implement a new feature. What was happening in the past is that you have the idea, you go to a UI/UX designer, and you ask him to prototype and visualize the idea. Now, with Claude and Claude Code, what you can do is just draw a picture like the one you see here while you're drinking a coffee in San Francisco, and then let Claude implement the idea for you. Let's see this in action. This is the Claude Code UI. You're familiar with the CLAUDE.md, which gives some instruction. Here we just say that we are a PM, and starting from the picture we want to render a prototype of the app, the wireframe that we'll then pass to the UX designer. In a few minutes, you can see how Claude was capable of rendering it starting from a very simple picture or drawing. Imagine how much time you save compared to the back and forth you were doing in the past to get this first prototype.
At this point, the PM creates a prototype and passes it to the UI/UX developer, who needs to implement a more solid interface to use in production. In this use case, we want to create at least three pages: from the landing to the thank-you message page and a dashboard that shows in real time the feedback I'll receive from the room. There are many ways to implement this, but here I want to use an additional capability of Claude Code, which is plan mode. With plan mode, we put Claude in a mode where it thinks and proposes what it's going to do before implementing any code. This is very important because it gives me a degree of freedom to change something according to my preference or some standard that I might access through an MCP server using Figma, for example. So we started from the wireframe from the PM with a similar prompt, I enabled plan mode, and in this case I'm simulating receiving some instruction from Figma using a design doc. Compared to before, it creates a plan on what it's going to do with respect to all the components defined in the slide. We look at them, we are happy, we accept, and Claude Code implements all of them. At the end, we get this optimized version. We shift from a prototype to something that can be used in this session in a very simple way.
This is the part that probably every one of you does every day. Let's wear the third hat, the software engineer. The software engineer receives this front end with all the components. Maybe it doesn't know how to deploy this application on Google Cloud. How can you have a clear picture of what components on Google Cloud you need to use in order to deploy a simple application like the one I show you today? Luckily, it's not a problem because, as Google Cloud, we invested a lot of time to integrate with this large vibe-coding ecosystem that is growing around models. In the last few months, we released two important components. The first is the Developer Knowledge API with its associated MCP server. The second is Google Cloud Skills. With the Developer Knowledge API, you get access to fresh documentation from Google Cloud that can be directly consumed in Claude Code through the MCP server, and it lets Claude Code figure out the best architecture and the best implementation to deploy a certain application on Google Cloud. This is important because it means you don't need to know how to deploy an application on Google Cloud — you can leverage Claude Code and this MCP server to build applications like this one.
In this case, we are going to deploy the feedback API on a serverless function like Cloud Run. We'll connect with a web-oriented DB like Firestore to collect the raw responses given through the feedback app. Then, because we want that analytics part, we'll build it so we can store those raw responses in an analytical data warehouse like BigQuery. We'll post-process and consume this information in a dashboard like the one you can find in Looker. You can build this using Claude Code in combination with the MCP server without any prior knowledge of how to do that. The MCP server and the Developer Knowledge API got paired with the skill part. If with the MCP server you design the architecture, with the skills you cover the single blocks of this architecture. For example, we released simple skills that enabled Claude to deploy an API on Cloud Run or to connect Cloud Run with Firestore. It's more about the implementation itself rather than giving the overall picture. Once you get Claude Code enabled with the MCP server of the documentation and the skills, you can directly implement the architecture, and you can do this in parallel. Another component you can use in Claude Code is sub agents. We are going to spin up three different sub agents — one for the API, one for the ingestion pipeline, and one for the dashboard — and you can parallelize the implementation of each of these components just like you run a team sprint in your normal development life cycle.
Let's see this in action. First, I want to show you that we have enabled the MCP server of the documentation and some of the skills we pre-built. Once more, I provide a very simple prompt. The first step is designing the cloud-native back end, so it provides me a draft of the architecture. In this case I could use plan mode again, but for simplicity I didn't. Then it used one of the skills I provided in order to implement the API. Let's say we are happy with the API spec; then we have the architecture and the API spec. The next step is running multiple agents in parallel in order to implement the three components of the app. It's pretty quick. As you notice, Claude Code also manages the testing part after it finishes the implementation, and at the end you get your app, which is now ready to deploy on Google Cloud.
At this point we have the code of the app ready to be deployed, but because we are deploying on cloud and we want to open this up to a larger audience, we want to deploy it confidently. This is when you want to consider running a security review. Depending on your company, you can have different security requirements. For example, you may want to check if your application is solid with respect to the most common OWASP issues, or because you're deploying on cloud, you will use what is called a service account and you want to limit the service account when it calls a particular API like the ones related to reading and writing to a DB. So you're sure you're limiting what can happen when the application operates on the cloud. These are just a couple of examples. This representation is a strong simplification of what can happen in real life — we're letting the security engineer not only approve if the app is secure enough to be deployed, but also deploy the app. This is just a demo, so we have that degree of freedom. Let's run the final demo. In this case I use a pre-built security review that you can find in Claude Code with a very simple prompt. Claude Code runs the first task, double-checks that everything is aligned, finds a possible issue, and automatically fixes it so the app is now secure. Once it is secure, you deploy the backend API. What we get at the end is an endpoint with our app. The app is live, so I'll quickly unlock my laptop and ask the backstage to share it. The app is up and running on GCP. For people that don't know GCP, this is Cloud Run, a serverless service you can use to deploy apps. Live, I can give a score — what do you think of the session so far? "Oh, five." Five? Thank you, man, I really appreciate that. In real time, it updates the number of responses, the score, and the visualization. Just for fun, I built a feedback analyzer. Once I click this, it calls Claude on Google Cloud, and based on the feedback and comments, it generates a summary.
At this point we have the app and we collected good feedback, but the development life cycle is still there. We have the last step to cover: people start using our app. One of the KPIs on the dashboard was the response time — how long it took you to provide feedback. You can collect this data, analyze it, and generate insights so they can be used to improve the application. If you are new on Google Cloud, there are several services you can use to analyze this data. One is an analytical data warehouse, which is BigQuery, and for the reporting part you have a tool like Looker. As we said before, you don't need to know how to use BigQuery or how to build a dashboard in Looker, because we also provide an MCP server for doing this. For the sake of time, I'm not going to demo how to use an MCP server to query BigQuery or build a dashboard, but I want to share where you can find this information after this session, because we are going to release the code.
With respect to all the MCP servers available on Google Cloud, we recently announced the agent platform. In the agent platform you have the agent registry, which has the list of all the MCP servers we natively support on Google Cloud. For example, we have the Developer Knowledge service I just showed you, and we also have the BigQuery MCP server that you can use to query the data we just collected. This registry is relevant because it tells you how to set up the MCP server on your side, gives you observability features, and the description of all the tools you'll find for the MCP server, so Claude Code is able to use this server to query your data. With respect to the Looker part, we released the MCP Toolbox for Databases. This is an open-source model context protocol server that includes an integration with Looker, and we have a very well-documented quick start on how to set it up with Claude Code and start using it to consume data from BigQuery and build your dashboard. I leave you this as an exercise. We are going to release the code, so you can go home and integrate these two parts. The dashboards you can create are pretty powerful.
Time to wrap up. What I tried to explain today is essentially two things. First, I was trying to show you how all the components of Claude Code, including skills, MCP servers, and sub agents, can really speed up the process of software development, as well as how you can use Claude Code with Claude models on GCP in a very seamless way. We ran several sessions across multiple personas, and the experience was just straightforward. This is what you can get if you combine Claude Code with Claude models on GCP. As I said, the code is going to be available right after the session. We have a great quick start and well-maintained documentation both on the Google Cloud side and the Anthropic side, so I highly recommend you go there and check it out. If you have questions or want to provide additional feedback, feel free to reach out. These are my social media accounts. With that being said, thank you so much, and it was a pleasure being here today.
TL;DR
- A Google Cloud developer advocate demos how to use Claude Code with Claude models on Google Cloud to build and deploy a feedback application end to end, role-playing five personas across the software development life cycle.
- Setting up Claude on Google Cloud is streamlined via Application Default Credentials (ADC) plus a wizard that detects your project, region, and available models — no API keys to rotate, per-token billing, provisioned throughput, and data staying in your project.
- The demo combines Claude Code features (plan mode, MCP servers, skills, sub agents, security review) with Google Cloud components (Developer Knowledge API, Cloud Run, Firestore, BigQuery, Looker) to ship a live app without prior GCP deployment expertise.
Takeaways
- Claude can augment every persona in the SDLC — PM, UI/UX designer, software engineer, security engineer, and data analyst — replacing slow back-and-forth handoffs.
- The simplest way to use Claude models on Google Cloud is Application Default Credentials, which auto-detect your credentials and environment; a setup wizard then detects project, region, and available models to pin.
- Reasons to run Claude on Google Cloud include per-token billing (no message cap), provisioned throughput for production, no API keys to rotate, project-scoped policies, data staying in your project, and multi-region (global/regional) endpoints.
- Plan mode lets Claude think and propose its approach before writing code, giving you a chance to adjust against standards (e.g. design specs pulled in via a Figma MCP server).
- Google Cloud's Developer Knowledge API (with its MCP server) feeds fresh GCP documentation into Claude Code so it can figure out the best architecture, while Google Cloud Skills cover concrete implementation blocks like deploying to Cloud Run or wiring Cloud Run to Firestore.
- Sub agents let you parallelize implementation — the demo spins up separate agents for the API, the ingestion pipeline, and the dashboard, mirroring a team sprint.
- The reference architecture deploys a feedback API on Cloud Run, stores raw responses in Firestore, post-processes data in BigQuery, and reports in Looker.
- A pre-built security review in Claude Code checks for common issues (e.g. OWASP risks, over-privileged service accounts), auto-fixes findings, and then deploys the secured backend.
Vocabulary
Application Default Credentials (ADC) — A Google Cloud authentication method that automatically discovers your credentials based on the environment, avoiding manual API key handling.
Provisioned throughput — Reserved model capacity on Google Cloud for enterprise applications that need predictable performance in production.
MCP server — A Model Context Protocol server that exposes tools and data (e.g. fresh documentation) to Claude Code.
Developer Knowledge API — A Google Cloud service that provides up-to-date GCP documentation consumable by Claude Code through an MCP server.
Skills — Pre-built capability blocks that teach Claude how to perform concrete implementation tasks, such as deploying an API on Cloud Run.
Sub agents — Parallel Claude agents that each own a component of the build, enabling concurrent implementation.
Plan mode — A Claude Code mode where the model proposes its plan before writing any code, allowing review and adjustment.
Cloud Run — A serverless Google Cloud service for deploying applications and APIs.
BigQuery — Google Cloud's analytical data warehouse used for post-processing and analyzing collected data.
Transcript
Hello everyone. Thank you for being here today. I'm pretty excited to have this session with you. I know we are almost at the end of the event, so thank you for bearing with us. To start, let me introduce myself. I'm Ivan Nardini. I'm a developer advocate at Google Cloud working on building content in partnership with Anthropic. To begin this presentation today, I want to ask a very simple question: how many of you in the last week used any AI tool to code or build an application? Okay, the majority. How many of you use the same AI tool to build and deploy applications on Google Cloud? Yeah, just a few. So the goal here today is just to try to make it better. In this demo, what I'm going to show you is how you can use Claude on Google Cloud to build and deploy an application end to end. To do that, I'm going to wear five different hats.
We will start by imagining the use case, the scenario. You are in an enterprise context. In an enterprise context, you probably have a team like the one we're visualizing here that is engaged to build a new feature or a new product. Starting from the left, you probably have a PM. The PM might have an idea of how to improve a particular product or how to implement a feature. Starting from this idea, he shares it with a UI/UX designer who allows him to design and visualize the idea. After the idea starts getting a shape, it is sent to a software engineer who starts developing the core logic behind it in order to ship and make the application accessible to everyone. Before shipping it, in order to be confident in what you are releasing, you probably have to pass a security review. That's why you have a role like the security engineer that allows you to be confident in the release. Finally, once the application gets released, you probably have a data persona like a growth marketer or a data analyst who analyzes the data collected through the app in order to generate insights and produce feedback to the PM so that the product or feature gets improved. These are all the personas that, in a simple way, we can imagine are involved in the software development life cycle. With respect to these personas, Claude — the Anthropic coding agent — can augment all of them, providing several components that we are going to use today.
For the remaining part of this presentation, I'm going to put on the hat of all these personas. I'm going to show you how you can use these Claude Code components together with Claude on Google Cloud in order to build and ship a very simple feedback application that, at the end of the presentation, we are going to use to rate my performance here. But before we start building, you need to set up Claude Code. I'm so proud that we worked together with Anthropic to make this process of setting up Claude Code to use models on Google Cloud very simple and straightforward. You have multiple methods to use Claude models on Google Cloud in Claude Code, but the simplest and fastest one is using Application Default Credentials, which automatically finds your credentials based on the environment that you're going to use. Recently, Claude Code also introduced a wizard that allows you to detect your project and your region where the models are served, check which models are available in your project, and let you pin them in order to start building your application.
At this point you're wondering, "Okay, but what's different about using Claude Code with Claude models? Why use Claude on GCP?" There are many reasons. First of all, you pay for what you use — the usage of Claude models on Google Cloud is per token, so you don't receive a message cap. Also, if you're building an enterprise application that needs to go to production, you can always access what is called provisioned throughput, which reserves some throughput for you. The other important reason is that the setup is straightforward using ADC. You don't have APIs to rotate or environment variables to set. You can access models in your project with your own policies set, and the data stays in your project while you are interacting with Claude Code. Models are served in multiple regions, so you have global endpoints and regional endpoints depending on the availability that you need. As Google Cloud, we have great availability service standards that allow you to use Claude on one of the most performing infrastructures in the market. These are some of the main reasons why you want to consider Claude on Google Cloud, especially in an enterprise context.
Now that you have a few reasons, we are ready to build. As I said, I'll start by wearing the hat of a PM. Imagine you want to improve our services or implement a new feature. What was happening in the past is that you have the idea, you go to a UI/UX designer, and you ask him to prototype and visualize the idea. Now, with Claude and Claude Code, what you can do is just draw a picture like the one you see here while you're drinking a coffee in San Francisco, and then let Claude implement the idea for you. Let's see this in action. This is the Claude Code UI. You're familiar with the CLAUDE.md, which gives some instruction. Here we just say that we are a PM, and starting from the picture we want to render a prototype of the app, the wireframe that we'll then pass to the UX designer. In a few minutes, you can see how Claude was capable of rendering it starting from a very simple picture or drawing. Imagine how much time you save compared to the back and forth you were doing in the past to get this first prototype.
At this point, the PM creates a prototype and passes it to the UI/UX developer, who needs to implement a more solid interface to use in production. In this use case, we want to create at least three pages: from the landing to the thank-you message page and a dashboard that shows in real time the feedback I'll receive from the room. There are many ways to implement this, but here I want to use an additional capability of Claude Code, which is plan mode. With plan mode, we put Claude in a mode where it thinks and proposes what it's going to do before implementing any code. This is very important because it gives me a degree of freedom to change something according to my preference or some standard that I might access through an MCP server using Figma, for example. So we started from the wireframe from the PM with a similar prompt, I enabled plan mode, and in this case I'm simulating receiving some instruction from Figma using a design doc. Compared to before, it creates a plan on what it's going to do with respect to all the components defined in the slide. We look at them, we are happy, we accept, and Claude Code implements all of them. At the end, we get this optimized version. We shift from a prototype to something that can be used in this session in a very simple way.
This is the part that probably every one of you does every day. Let's wear the third hat, the software engineer. The software engineer receives this front end with all the components. Maybe it doesn't know how to deploy this application on Google Cloud. How can you have a clear picture of what components on Google Cloud you need to use in order to deploy a simple application like the one I show you today? Luckily, it's not a problem because, as Google Cloud, we invested a lot of time to integrate with this large vibe-coding ecosystem that is growing around models. In the last few months, we released two important components. The first is the Developer Knowledge API with its associated MCP server. The second is Google Cloud Skills. With the Developer Knowledge API, you get access to fresh documentation from Google Cloud that can be directly consumed in Claude Code through the MCP server, and it lets Claude Code figure out the best architecture and the best implementation to deploy a certain application on Google Cloud. This is important because it means you don't need to know how to deploy an application on Google Cloud — you can leverage Claude Code and this MCP server to build applications like this one.
In this case, we are going to deploy the feedback API on a serverless function like Cloud Run. We'll connect with a web-oriented DB like Firestore to collect the raw responses given through the feedback app. Then, because we want that analytics part, we'll build it so we can store those raw responses in an analytical data warehouse like BigQuery. We'll post-process and consume this information in a dashboard like the one you can find in Looker. You can build this using Claude Code in combination with the MCP server without any prior knowledge of how to do that. The MCP server and the Developer Knowledge API got paired with the skill part. If with the MCP server you design the architecture, with the skills you cover the single blocks of this architecture. For example, we released simple skills that enabled Claude to deploy an API on Cloud Run or to connect Cloud Run with Firestore. It's more about the implementation itself rather than giving the overall picture. Once you get Claude Code enabled with the MCP server of the documentation and the skills, you can directly implement the architecture, and you can do this in parallel. Another component you can use in Claude Code is sub agents. We are going to spin up three different sub agents — one for the API, one for the ingestion pipeline, and one for the dashboard — and you can parallelize the implementation of each of these components just like you run a team sprint in your normal development life cycle.
Let's see this in action. First, I want to show you that we have enabled the MCP server of the documentation and some of the skills we pre-built. Once more, I provide a very simple prompt. The first step is designing the cloud-native back end, so it provides me a draft of the architecture. In this case I could use plan mode again, but for simplicity I didn't. Then it used one of the skills I provided in order to implement the API. Let's say we are happy with the API spec; then we have the architecture and the API spec. The next step is running multiple agents in parallel in order to implement the three components of the app. It's pretty quick. As you notice, Claude Code also manages the testing part after it finishes the implementation, and at the end you get your app, which is now ready to deploy on Google Cloud.
At this point we have the code of the app ready to be deployed, but because we are deploying on cloud and we want to open this up to a larger audience, we want to deploy it confidently. This is when you want to consider running a security review. Depending on your company, you can have different security requirements. For example, you may want to check if your application is solid with respect to the most common OWASP issues, or because you're deploying on cloud, you will use what is called a service account and you want to limit the service account when it calls a particular API like the ones related to reading and writing to a DB. So you're sure you're limiting what can happen when the application operates on the cloud. These are just a couple of examples. This representation is a strong simplification of what can happen in real life — we're letting the security engineer not only approve if the app is secure enough to be deployed, but also deploy the app. This is just a demo, so we have that degree of freedom. Let's run the final demo. In this case I use a pre-built security review that you can find in Claude Code with a very simple prompt. Claude Code runs the first task, double-checks that everything is aligned, finds a possible issue, and automatically fixes it so the app is now secure. Once it is secure, you deploy the backend API. What we get at the end is an endpoint with our app. The app is live, so I'll quickly unlock my laptop and ask the backstage to share it. The app is up and running on GCP. For people that don't know GCP, this is Cloud Run, a serverless service you can use to deploy apps. Live, I can give a score — what do you think of the session so far? "Oh, five." Five? Thank you, man, I really appreciate that. In real time, it updates the number of responses, the score, and the visualization. Just for fun, I built a feedback analyzer. Once I click this, it calls Claude on Google Cloud, and based on the feedback and comments, it generates a summary.
At this point we have the app and we collected good feedback, but the development life cycle is still there. We have the last step to cover: people start using our app. One of the KPIs on the dashboard was the response time — how long it took you to provide feedback. You can collect this data, analyze it, and generate insights so they can be used to improve the application. If you are new on Google Cloud, there are several services you can use to analyze this data. One is an analytical data warehouse, which is BigQuery, and for the reporting part you have a tool like Looker. As we said before, you don't need to know how to use BigQuery or how to build a dashboard in Looker, because we also provide an MCP server for doing this. For the sake of time, I'm not going to demo how to use an MCP server to query BigQuery or build a dashboard, but I want to share where you can find this information after this session, because we are going to release the code.
With respect to all the MCP servers available on Google Cloud, we recently announced the agent platform. In the agent platform you have the agent registry, which has the list of all the MCP servers we natively support on Google Cloud. For example, we have the Developer Knowledge service I just showed you, and we also have the BigQuery MCP server that you can use to query the data we just collected. This registry is relevant because it tells you how to set up the MCP server on your side, gives you observability features, and the description of all the tools you'll find for the MCP server, so Claude Code is able to use this server to query your data. With respect to the Looker part, we released the MCP Toolbox for Databases. This is an open-source model context protocol server that includes an integration with Looker, and we have a very well-documented quick start on how to set it up with Claude Code and start using it to consume data from BigQuery and build your dashboard. I leave you this as an exercise. We are going to release the code, so you can go home and integrate these two parts. The dashboards you can create are pretty powerful.
Time to wrap up. What I tried to explain today is essentially two things. First, I was trying to show you how all the components of Claude Code, including skills, MCP servers, and sub agents, can really speed up the process of software development, as well as how you can use Claude Code with Claude models on GCP in a very seamless way. We ran several sessions across multiple personas, and the experience was just straightforward. This is what you can get if you combine Claude Code with Claude models on GCP. As I said, the code is going to be available right after the session. We have a great quick start and well-maintained documentation both on the Google Cloud side and the Anthropic side, so I highly recommend you go there and check it out. If you have questions or want to provide additional feedback, feel free to reach out. These are my social media accounts. With that being said, thank you so much, and it was a pleasure being here today.