mcp-vertex-ai-retail-search-server

ksmin23/mcp-vertex-ai-retail-search-server

3.2

If you are the rightful owner of mcp-vertex-ai-retail-search-server and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to henry@mcphub.com.

This project is a Model Context Protocol (MCP) server example that provides the Vertex AI Search for Commerce API from Google Cloud as a tool, using FastMCP.

Tools

  1. search_products

    Searches for products in the product catalog based on a given query.

Vertex AI Search for Commerce MCP Server

This project is a Model Context Protocol (MCP) server example that provides the Vertex AI Search for Commerce API from Google Cloud as a tool, using FastMCP.

Through this server, an AI agent can search a product catalog using natural language queries.

Key Features

  • Lightweight MCP server based on FastMCP
  • Provides Vertex AI Search for Commerce product search functionality (search_products)
  • Easy setup using a .env file

Prerequisites

  • Python 3.8 or higher
  • uv (a fast Python package installer and virtual environment manager)
  • Google Cloud CLI (gcloud)

Installation and Setup

  1. Google Cloud Authentication

    Set up Application Default Credentials (ADC) to access Google Cloud services from your local environment. Run the command below in your terminal and complete the authentication through your browser.

    gcloud auth application-default login

  2. Environment Variable Setup

    Copy the .env.example file in the project root directory to a new .env file, then modify the contents to match your Google Cloud environment.

    cp .env.example .env

    .env file contents:

    PROJECT_ID="your-gcp-project-id"
LOCATION="global"
CATALOG_ID="default_catalog"
SERVING_CONFIG_ID="default_serving_config"

  3. Create Virtual Environment and Install Dependencies

    Use uv to create a virtual environment and install the necessary libraries.

    # Create virtual environment
uv venv

# Activate virtual environment (macOS/Linux)
source .venv/bin/activate
# (Windows: .venv\Scripts\activate)

# Install dependencies
uv pip install -r requirements.txt

Running the Server

You can run the Python script directly using uv's execution feature.

uv run python src/server.py

Running the Server (using fastmcp)

You can also run the server using the CLI provided by the fastmcp library. This method automatically finds and runs the FastMCP instance.

Specify the module path where the mcp instance is located. You can use the --port option to specify a different port instead of the default (8080).

fastmcp run src/server.py --transport http --port 9000

When the server starts successfully, you will see a message like this:

INFO:     Started server process [12345]
INFO:     Waiting for application startup.
INFO:     Application startup complete.
INFO:     Uvicorn running on http://127.0.0.1:9000 (Press CTRL+C to quit)

Provided Tools

This MCP server provides the following tool:

search_products

  • Description: Searches for products in the product catalog based on a given query.
  • Parameters:
    • query (str): The product keyword to search for (e.g., "jeans", "sneakers").
    • visitor_id (str, optional): A unique ID to identify the user. Used for personalized search results. (Default: "guest-user")
  • Returns: A list of searched products (each product is a dictionary containing id, title, price, and uri information).

Google Cloud Run Deployment Guide

This section guides you on how to deploy the MCP server to Google Cloud Run.

1. Prerequisites

This setup is for interacting with Google Cloud from your local machine.

  • Install Google Cloud SDK: If you don't have the gcloud CLI installed, refer to the official documentation to install it.

  • gcloud Authentication:

    gcloud auth login

  • Set Google Cloud Project:

    gcloud config set project [YOUR_PROJECT_ID]

    (Replace [YOUR_PROJECT_ID] with your actual GCP project ID.)

  • Enable Required APIs:

    gcloud services enable run.googleapis.com \
    artifactregistry.googleapis.com

  • Configure Docker Authentication:

    gcloud auth configure-docker [REGION]-docker.pkg.dev

    (Replace [REGION] with a GCP region like asia-northeast3.)

2. Create a Dockerfile

Create a Dockerfile in the project root. This file defines how to containerize the application.

# 1. Set the base image
FROM python:3.11-slim

# 2. Set environment variables
ENV PYTHONDONTWRITEBYTECODE 1
ENV PYTHONUNBUFFERED 1

# 3. Set the working directory
WORKDIR /app

# 4. Install dependencies
COPY requirements.txt .
RUN pip install uv && uv pip install --no-cache -r requirements.txt

# 5. Copy source code
COPY . .

# 6. Expose the port
EXPOSE 8080

# 7. Run the application
# Use the fastmcp CLI to run the server in a production environment.
CMD ["fastmcp", "run", "src/server.py", "--transport", "http", "--host", "0.0.0.0", "--port", "8080"]

Note: Since we are using fastmcp, ensure that fastmcp is included in your requirements.txt.

3. Create an Artifact Registry Repository

Create a repository in Artifact Registry to store your Docker images.

gcloud artifacts repositories create [REPOSITORY_NAME] \
    --repository-format=docker \
    --location=[REGION] \
    --description="MCP Search Server repository"
  • [REPOSITORY_NAME]: Specify a desired repository name, such as mcp-repo.
  • [REGION]: Use the same region as in the previous step.

4. Build and Push the Image

There are two ways to build the container image and push it to Artifact Registry.

Method 1: Using Local Docker

This method requires Docker to be installed on your local machine.

# 1. Build the image
docker build -t [REGION]-docker.pkg.dev/[YOUR_PROJECT_ID]/[REPOSITORY_NAME]/mcp-vertexai-retail-search-server:latest .

# 2. Push the image
docker push [REGION]-docker.pkg.dev/[YOUR_PROJECT_ID]/[REPOSITORY_NAME]/mcp-vertexai-retail-search-server:latest

(Replace [REGION], [YOUR_PROJECT_ID], and [REPOSITORY_NAME] with your actual values.)

Method 2: Using Google Cloud Build (Recommended)

This method does not require a local Docker installation and uses Google Cloud's managed build service for faster and more reliable image building and pushing.

Run the following single command from your project root directory:

gcloud builds submit --tag [REGION]-docker.pkg.dev/[YOUR_PROJECT_ID]/[REPOSITORY_NAME]/mcp-vertexai-retail-search-server:latest .

(Replace [REGION], [YOUR_PROJECT_ID], and [REPOSITORY_NAME] with your actual values.)

This command sends the source code from the current directory to Cloud Build, builds the image using the Dockerfile, and saves it to Artifact Registry with the specified tag, automating the entire process.

5. Deploy to Cloud Run

When deploying to Cloud Run, it's a best practice to use a dedicated service account to grant the service the minimum permissions required. This enhances security by following the principle of least privilege.

Using a Service Account

  1. Create a Service Account: First, create a new service account for your Cloud Run service.

    gcloud iam service-accounts create mcp-vaisc-sa \
    --display-name="MCP Vertex AI Search for Commerce Service Account"
    • mcp-vaisc-sa: This is the ID of the service account. You can change it to your desired name.

  2. Grant Permissions: The service account needs permissions to access the Vertex AI Search for Commerce API. Grant the Retail Viewer role to the service account, which provides the necessary read-only permissions for searching products.

    gcloud projects add-iam-policy-binding [YOUR_PROJECT_ID] \
    --member="serviceAccount:mcp-vaisc-sa@[YOUR_PROJECT_ID].iam.gserviceaccount.com" \
    --role="roles/retail.viewer"
    • Replace [YOUR_PROJECT_ID] with your actual GCP project ID.

Now, you can deploy the service with this service account.

You can deploy the service in two ways: publicly accessible or restricted to a VPC.

Public Deployment

The following command deploys the service to be publicly accessible. The --ingress all setting is default, and --allow-unauthenticated permits public access.

gcloud run deploy mcp-vaisr-server \
    --image [REGION]-docker.pkg.dev/[YOUR_PROJECT_ID]/[REPOSITORY_NAME]/mcp-vertexai-retail-search-server:latest \
    --region [REGION] \
    --service-account "mcp-vaisc-sa@[YOUR_PROJECT_ID].iam.gserviceaccount.com" \
    --allow-unauthenticated
  • --allow-unauthenticated: This flag allows anyone to access the service. If authentication is required, remove this flag.
Secure Deployment within a VPC

To deploy the service so that it is only accessible from within a specific VPC network for enhanced security, use the deploy_to_cloud_run.py script. This script sets the ingress settings to internal, allowing access only from the specified VPC network.

python deploy_to_cloud_run.py --service-name internal-mcp-vaisr-server \
    --network [VPC] \
    --subnet [SUBNET] \
    --ingress internal \
    --vpc-egress all-traffic \
    --service-account "mcp-vaisc-sa@[YOUR_PROJECT_ID].iam.gserviceaccount.com"

For more details on Cloud Run ingress settings, refer to the official documentation.

Once the deployment is complete, you can access your application via the provided service URL.

Reference