Challenge 3: Talking to BigQuery
Previous Challenge Next Challenge
Introduction
Sara’s goal is to analyze how the new banking product is performing, which requires querying a company data store. Since she is a product owner, not a database administrator, she wants to use natural language instead of SQL.
Writing custom code to map user queries to database schemas can be incredibly tedious. Instead, we’ll let our model generate the SQL queries and use a tool to access the underlying data source and run queries.
We could build our own tool as we did in the previous challenge, but there’s also a plethora of tools available built by others. This is where the Model Context Protocol (MCP) plays a role; it offers a standardized abstraction layer for tools so that any agent can use them.
For this challenge we’ll use the the Google-managed BigQuery MCP Server to access the company data source with customer data.
Description
Integrate the BigQuery MCP Server into your ADK agent. Once the agent is equipped with the BigQuery MCP tools, and can run SQL queries successfully, commit and push your changes.
Success Criteria
- Ask the agent: How many accounts were created in the last quarter?. This should successfully retrieve the result from BigQuery (around 150, exact numbers might be different to randomly generated data).
- Verify the agent utilizes the MCP tools to inspect and query the database under the hood.
- All the changes are committed and pushed to the Git repository.
Tips
- Keep in mind that this MCP server is a remote server available through Streamable HTTP.