Use AI assistance for Data Connect

Use Gemini in Firebase to help craft schemas, queries, and mutations for your client-side code.

Describe your app, data model, or a desired query or mutation in natural language, and Gemini in Firebase generates the Data Connect equivalent.

This AI assistance is available in the following development contexts:

In the Firebase console, you can generate, test, and deploy your schemas and operations.
In your local environment, you can use the Firebase CLI and Data Connect VS Code extension to generate, test, and develop on your app with the emulator.
AI-powered development tools can use Firebase MCP server to generate, test, develop your app.

Learn more about Data Connect schema , query and mutation syntax in guides.

How AI assistance for Data Connect uses your data

For more information, see How Gemini in Firebase uses your data .

Set up AI assistance for Data Connect

To use AI assistance with Data Connect , enable Gemini in Firebase as described in Set up Gemini in Firebase .

Generate schema, queries, and mutations with Gemini in Firebase

You can use AI assistance for Data Connect in many of your workflows.

In the Firebase console

When you create a Data Connect service, the Firebase console offers a "Getting started with Gemini" experience.

You can describe an app idea, and AI assistance generates the following:

A complete schema based on your app idea.
Example operations and data mutations.

From the data page, you can use the Help me write GraphQL pen_spark button to generate and execute operations based on natural language. Check out some example use cases .

This workflow is described in our Get started guide . You can continue in your local development environment with the deployed schema and operations.

In your local environment

You can also get AI assistance from the Firebase CLI and the Data Connect VS Code extension.

You can provide your app idea to firebase init dataconnect , and it generates the following:

A complete schema based on your app idea.
Example operations and a seed data mutation.

The Data Connect VS Code extension provides the following features:

Generates/Refine Operations Code Lensto convert GraphQL comments into Data Connect operations.
Seamless integration with Gemini Code Assist and Firebase MCP server.

This workflow is described in our Get started guide for local prototyping .

Use the Firebase MCP server with AI-powered development tools

The Firebase MCP server works with any AI assistant tools that can act as an MCP client, including Gemini CLI, Gemini Code Assist , Cursor, Visual Studio Code Copilot, Claude Desktop, and Windsurf Editor.

The Firebase MCP server provides additional context and capabilities to help AI-powered development tools work better with Data Connect . It can do the following:

Set up new project directories and generated SDKs.
Build and iterate on schemas, operations based on compile errors.
Generate new schemas, operations based on requirements.
Generate and execute operations against local emulator or backend services.
Gather information on existing services.

To use Firebase MCP server:

Set up your MCP client following this guide .
Ask for help related to Data Connect . Examples prompts:
1. "Set up a Data Connect project for a pizza delivery app."
2. "Fix Data Connect compile errors."
3. "In the home page, I need to show active chat rooms and friend list. Generate a Data Connect query."
4. "What users are in my local Data Connect emulator?"
5. "What Google Cloud regions are my Data Connect services in?"

Example use cases for generating operations

The following sections describe sample use cases:

Return the top five movies in descending order by rating
Create a mutation that adds a movie to the database based on user input
Create a query that lists reviews based on user-provided genre and ratings

Return the top five movies in descending order by rating

To use AI assistance for Data Connect to generate GraphQL based on natural language:

Open Data Connect in your project and, under Services, select your data source.
Click Data.

Click the Help me write GraphQL pen_spark icon. Describe in natural language the query or mutation you want to generate, and click Generate.

For example, if you're using the Movies data source referenced in the "Build with Data Connect (web)" codelab , you could ask, " Return the top five movies of 2022, in descending order by rating," which might return a result like the following:

  query 
  
 TopMovies2022 
  
 { 
  
 movies 
 ( 
 where 
 : 
  
 { 
 releaseYear 
 : 
  
 { 
 eq 
 : 
  
 2022 
 }}, 
  
 orderBy 
 : 
  
 [{ 
 rating 
 : 
  
 DESC 
 }], 
  
 limit 
 : 
  
 5 
 ) 
  
 { 
  
 id 
  
 title 
  
 rating 
  
 releaseYear 
  
 } 
 }

Review the response:
- If the response looks correct, click Insertto insert the response into the code editor.
- If the response could be refined, click Edit, update the prompt, and click Regenerate.
After you've accepted the response, set the following in the Parameterssection, if applicable:
- Variables: If your query or mutation contains variables, define them here. Use JSON to define them, for example, {"title":"The Matrix", "releaseYear":"1999"} .
- Authorization: Choose the Authorization context (Administrator, Authenticated, or Unauthenticated) with which to run the query or mutation.
Warning: If you're running a mutation , review it carefully before running it. Running a mutation in the data editor will execute the request.
Click Runin the code editor and review results.

To test multiple queries or mutations in the code editor, ensure they are named. For example, the following query is named GetMovie . Move your cursor into the first line of the query or mutation to activate the Runbutton.

  query 
  
 GetMovie 
 ( 
 $myKey 
 : 
  
 Movie_Key 
 !) 
  
 { 
  
 movie 
 ( 
 key 
 : 
  
 $myKey 
 ) 
  
 { 
  
 title 
  
 } 
 }

Create a mutation that adds a movie to the database based on user input

This example shows how to use natural language to generate a GraphQL mutation that populates your database. This example assumes that you're using the movie database schema used in the Firebase Data Connect documentation and "Build with Data Connect (web)" codelab .

From the Firebase console , open Data Connect .
Select your service and data source, then open the Datatab.

Click the Help me write GraphQL pen_spark icon and describe your mutation:

 Create  
a  
movie  
based  
on  
user  
input.

Click Generate. The mutation is returned. For example, Gemini might return a mutation like:

  mutation 
  
 CreateMovie 
 ( 
 $title 
 : 
  
 String 
 !, 
  
 $releaseYear 
 : 
  
 Int 
 !, 
  
 $genre 
 : 
  
 String 
 !, 
  
 $rating 
 : 
  
 Float 
 !, 
  
 $description 
 : 
  
 String 
 !, 
  
 $imageUrl 
 : 
  
 String 
 !, 
  
 $tags 
 : 
  
 [ 
 String 
 !] 
  
 = 
  
 []) 
  
 @ 
 auth 
 ( 
 level 
 : 
  
 USER 
 ) 
  
 { 
  
 movie_insert 
 ( 
 data 
 : 
  
 { 
  
 title 
 : 
  
 $title 
 , 
  
 releaseYear 
 : 
  
 $releaseYear 
 , 
  
 genre 
 : 
  
 $genre 
 , 
  
 rating 
 : 
  
 $rating 
 , 
  
 description 
 : 
  
 $description 
 , 
  
 imageUrl 
 : 
  
 $imageUrl 
 , 
  
 tags 
 : 
  
 $tags 
  
 }) 
 }

Review the output. If needed, click Editto refine the prompt and click Regenerate.
Next, click Insertto insert the mutation into the data editor.

To execute the mutation, you'll need to add variables. From the Parameterssection, open Variablesand include some test variables:

  { 
 " 
 title 
 ":" 
 My 
  
 amazing 
  
 movie 
 " 
 , 
  
 " 
 releaseYear 
 ":2024 
 , 
  
 " 
 genre 
 ": 
  
 " 
 Comedy 
 " 
 , 
 " 
 rating 
 ": 
  
 8 
 , 
  
 " 
 description 
 ": 
  
 " 
 A 
  
 new 
  
 movie 
  
 to 
  
 test 
  
 mutations 
 " 
 , 
 " 
 imageUrl 
 ": 
  
 "" 
 , 
  
 " 
 tags 
 ": 
  
 [" 
 comedy 
 " 
 , 
 " 
 space 
  
 travel 
 "] 
 }

Click Run.

Tip: You can also ask Gemini to expand the output. For example, to include comments and your source prompt in the output, you might use "Create a movie based on user input. Be sure to document the mutation and include this prompt in your response".

Next, create a query that verifies that your movie was added. Click the Help me write GraphQL pen_spark and, in the box that appears, type your prompt:

 List  
all  
movies  
from  
 2024 
  
that  
include  
all  
of  
these  
tags:  
 'space travel' 
,  
 'comedy' 
.

Gemini might return a response like the following:

  query 
  
 ComedySpaceTravelMovies2024 
  
 @ 
 auth 
 ( 
 level 
 : 
  
 PUBLIC 
 ) 
  
 { 
  
 movies 
 ( 
  
 where 
 : 
  
 { 
  
 releaseYear 
 : 
  
 { 
  
 eq 
 : 
  
 2024 
  
 }, 
  
 tags 
 : 
  
 { 
  
 includesAll 
 : 
  
 [ 
 "space travel" 
 , 
  
 "comedy" 
 ] 
  
 } 
  
 } 
  
 ) 
  
 { 
  
 id 
  
 title 
  
 imageUrl 
  
 releaseYear 
  
 genre 
  
 rating 
  
 description 
  
 tags 
  
 } 
 }

Insert and run the query. The movie you added should appear in the Historyfield.

Create a query that lists reviews based on user-provided genre and ratings

This example shows how to use natural language to generate a GraphQL query. This example assumes that you're using the movie database used in the Firebase Data Connect documentation and "Build with Data Connect (web)" codelab .

From the Firebase console , open Data Connect .
Select your service and data source, then open the Datatab.

Click the Help me write GraphQL pen_spark icon and describe your query:

 List  
all  
movie  
reviews,  
based  
on  
user-configurable  
genre  
and  
ratings.

Click Generate. The query is returned. For example, Gemini might return a query like:

  query 
  
 ListReviewsByGenreAndRating 
 ( 
 $genre 
 : 
  
 String 
 , 
  
 $minRating 
 : 
  
 Int 
 , 
  
 $maxRating 
 : 
  
 Int 
 ) 
  
 @ 
 auth 
 ( 
 level 
 : 
  
 PUBLIC 
 ) 
  
 { 
  
 reviews 
 ( 
 where 
 : 
  
 { 
  
 movie 
 : 
  
 { 
  
 genre 
 : 
  
 { 
 eq 
 : 
  
 $genre 
 } 
  
 }, 
  
 rating 
 : 
  
 { 
 ge 
 : 
  
 $minRating 
 , 
  
 le 
 : 
  
 $maxRating 
 } 
  
 }) 
  
 { 
  
 id 
  
 user 
  
 { 
  
 username 
  
 } 
  
 movie 
  
 { 
  
 title 
  
 genre 
  
 } 
  
 rating 
  
 reviewText 
  
 reviewDate 
  
 } 
 }

Review the output. If needed, click Editto refine the prompt and click Regenerate.
Next, click Insertto insert the mutation into the data editor.

To test this query, you'll need to add variables. From the Parameterssection, open Variablesand include variables to use for testing:

  { 
 "genre" 
 : 
 "sci-fi" 
 , 
  
 "minRating" 
 : 
 4 
 , 
  
 "maxRating" 
 : 
 9 
 }

Click Run.

Design prompts for third-party AI assistance tools

As with all AI assistance tools, better prompts yield more useful outputs.

When you provide natural language prompts to Gemini in Firebase , behind the scenes, the assistant translates your inputs to a more fully-developed prompt.

If you're working with third-party AI tools like Cursor or Windsurf, you can get better Data Connect recommendations by using similar, more detailed prompts.

We've published prompt templates for you to download, adapt, and copy into your IDE:

A template prompt for schema generation
A template prompt for operation generation

After downloading and modifying, create a prompt in familiar tooling (for example Cursor or Windsurf) as follows:

In Cursor (be sure to review the latest instructions from Cursor ):
1. Click the settings icon on top right.
2. Select the Rulestab.
3. Under the Project Rules, click Add a new rulebutton.
4. Copy and paste the rule.
In Windsurf (be sure to review the latest instructions from Windsurf ):
1. Open the Cascade window by clicking the Cascadebutton in the top right corner.
2. Click the Customizationsicon in the top right slider menu in Cascade, then navigate to the Rulespanel.
3. Click the + Globalor + Workspacebutton to create new rules at either the global or workspace level, respectively.
4. Copy and paste the rule.

Troubleshoot AI assistance for Data Connect

See Troubleshoot Gemini in Firebase .

Pricing

AI assistance for Data Connect is available as part of Gemini in Firebase and is included for individual users.

See Gemini in Firebase pricing for more information.

Next steps

Learn more about schema , query and mutation .
Learn more about Gemini in Firebase .