Build contexts using Gemini CLI

Preview — Data agents

This feature is subject to the "Pre-GA Offerings Terms" in the General Service Terms section of the Service Specific Terms, and the Additional Terms for Generative AI Preview Products. Pre-GA features are available "as is" and might have limited support. For more information, see the launch stage descriptions.

For information about access to this release, see the access request page.

This document describes how to use the Gemini CLI and the MCP toolbox to build agent context files. These files contain templates and facets that provide context for generating SQL queries from natural language. You will also use the DB Context Enrichment MCP Server.

To learn about data agents, see Data agents overview.

To build agent context, perform the following high-level steps:

• Prepare your environment
• Generate targeted templates
• Generate targeted facets
• Optional. Generate bulk templates

Before you begin

Complete the following prerequisites before creating an agent.

Prepare an AlloyDB for PostgreSQL cluster and instance

• Make sure that you have access to an existing AlloyDB cluster and instance or create a new one.
• Populate the database with the data and schema that the end user wants to access.

Required roles and permissions

• Add an Identity and Access Management (IAM) user or service account to the cluster at the database level. For more information, see Manage database users.
• Grant the alloydb.databaseUser role to the IAM user at the project level. For more information, see Add IAM policy binding for a project.

Prepare your environment

You can build agent context files from any any local development environment or IDE. To prepare the environment, perform the following steps:

• Install Gemini CLI
• Install and setup MCP toolbox
• Install and setup the DB Context Enrichment MCP Server

Install Gemini CLI

To install Gemini CLI, see Get Started with Gemini CLI. Make sure that you install Gemini CLI in a separate directory, which is also used to install the MCP toolbox and the DB Context Enrichment MCP Server.

Install and setup MCP toolbox

1. In the same directory where you installed Gemini CLI, install the MCP Toolbox Gemini CLI extension:

   gemini extensions install https://github.com/gemini-cli-extensions/mcp-toolbox

2. Create a tools.yaml configuration file in the same directory where you installed the MCP toolbox for configuring the database connection:

     sources:
       alloydb-pg-source:
         kind: alloydb-postgres
         project: PROJECT_ID
         region: REGION_ID
         cluster: CLUSTER_ID
         instance: INSTANCE_ID
         database: DATABASE_ID
         user: USER_NAME
         password: PASSWORD
   

   Replace the following:

   • PROJECT_ID: Your Google Cloud project ID.
   • REGION_ID: The region of your AlloyDB cluster (e.g., us-central1).
   • CLUSTER_ID: The ID of your AlloyDB cluster.
   • INSTANCE_ID: The ID of your AlloyDB primary instance.
   • DATABASE_ID: The name of the database to connect to.
   • USER_NAME: The database user. For more information about how to set this value, see Sources in MCP toolbox.
   • PASSWORD: The password for the database user. For more information about how to set this value, see Sources in MCP toolbox.

   For more information, see Quickstart (MCP with AlloyDB).

3. Verify that the tools.yaml file is configured correctly:

   ./toolbox --tools-file "tools.yaml"

Install DB Context Enrichment MCP Server

The DB Context Enrichment MCP Server provides a guided, interactive workflow to generate structured NL2SQL templates from your database schemas. It relies on the MCP Toolbox extension for database connectivity. For more information about installing the DB Context Enrichment MCP Server, see DB Context Enrichment MCP Server.

To install the DB Context Enrichment MCP Server, do the following:

1. In the same directory where you installed Gemini CLI, install uv Python package installer.

     pip install --user pipx
     pipx ensurepath
     pipx install uv

2. Install the DB Context Enrichment MCP Server.

   gemini extensions install https://github.com/GoogleCloudPlatform/db-context-enrichment

The server uses Gemini API for generation. Make sure that you export your API key as an environment variable. For more information about how to find your API key, see Using Gemini API keys.

To export the Gemini API key, run the following command:

export GEMINI_API_KEY="YOUR_API_KEY"

Replace YOUR_API_KEY with your Gemini API key.

Generate targeted templates

If you want to add a specific query pair as a query template to the agent context, then you can use the /generate_targeted_templates command. For more information about templates, see Data agents overview.

To add a query template to the agent context, perform the following steps:

1. In the same directory where you installed Gemini CLI, start Gemini:

   gemini

2. Complete the Gemini CLI Authentication Setup.

3. Verify that the MCP toolbox and the database enrichment extension are ready to use:

   /mcp list

4. Run the /generate_targeted_templates command:

   /generate_targeted_templates

5. Enter the natural language query that you want to add to the query template.

6. Enter the corresponding SQL query to the query template.

7. Review the generated query template. You can either save the query template as an agent context file or append it to an existing context file.

The agent context file similar to my-cluster-psc-primary_postgres_templates_20251104111122.json is saved in the directory where you ran the commands.

For more information about the context file and the query template, see Agent context.

Generate targeted facets

If you want to add a specific query pair as a facet to the agent context file, then you can use the /generate_targeted_fragments command. For more information about facets, see Data agents overview.

To add a facet to the agent context, perform the following steps:

1. Run the /generate_targeted_fragments command:

   /generate_targeted_fragments

2. Enter the natural language query that you want to add to the query template.

3. Enter the corresponding SQL query to the query template.

4. Review the generated facet. You can either save the facet to an agent context file or append it to an existing context file.

The agent context file similar to my-cluster-psc-primary_postgres_templates_20251104111122.json is saved in the directory where you ran the commands.

For more information about the context file and facets, see Agent context.

Optional: Generate bulk templates

If you want to auto-generate the agent context file based on your database schema and data, then you can use the /generate_bulk_templates command.

To auto-generate bulk templates, perform the following steps:

1. Run the /generate_bulk_templates command:

   /generate_bulk_templates

2. Based on your database schema, the template-based SQL generation takes you through a series of questions related to verifying the database information and granting permissions to access the database schema.

3. Review the generated query template. You can either approve the template or update a query pair that you want to revise.

4. Enter the natural language query that you want to add to the query template.

5. Enter the corresponding SQL query to the query template.

6. Review the generated query template. You can either save the query template as an agent context file or append it to an existing context file.

7. After you approve the query template, you can either create a new template file or append the query pairs to an existing template file. The query template is saved as a JSON file in your local directory.

The agent context file similar to my-cluster-psc-primary_postgres_templates_20251104111122.json is saved in the directory where you ran the commands.

For more information about the agent context file, see Agent context.

What's next

• Learn more about data agents.
• Learn how to create or delete a data agent in AlloyDB Studio.
• Learn how to inspect and call a data agent.