The Model Context Protocol (MCP) extends the capabilities of AI tools used by millions of developers every day. By connecting to MCP Servers, coding agents can directly integrate with additional tools and services in a standardized fashion.

We’re launching an experimental Firebase MCP Server as part of the Firebase CLI to provide AI assistants access to powerful tools for configuring and working with Firebase. By installing this locally you can integrate it with many popular IDEs and LLM clients.

What you can do with the Firebase MCP Server

The Firebase MCP server provides tools for managing, building, and observing your app with Firebase. It includes over 30 tools to assist your development and automatically activates the tools for your specific project.

With the Firebase MCP server, your AI tools can:

• Initialize a project directory, create projects/apps, and download important configuration like client SDK config.

• Manage users in your app including adding custom claims, looking up users by email, and listing users.

• Explore your Firestore database including reading/writing documents, listing collections, and querying.

• Execute GraphQL queries for Data Connect.

• Read and validate security rules for Firestore and Storage.

• Send messages to topics and devices using Cloud Messaging.

• Deploy or roll back a Remote Config template.

• Understand the top crashes impacting your application in production.

• Generate download URLs for Cloud Storage.

You can see the full list of available tools in our documentation.

We’ll continue to refine the capabilities and release updates as they become available. But please remember, The Firebase MCP Server is considered “experimental”. We may make significant changes before its stable release!

Getting Started with the Firebase MCP Server

For most MCP clients, you can add the Firebase MCP Server to your configuration file using this snippet:

{
  "mcpServers": {
    "firebase": {
      "command": "npx",
      "args": ["-y", "firebase-tools@latest", "experimental:mcp"]
    }
  }
}

This will download and run the latest version of the Firebase MCP Server and remain automatically up-to-date as we continue to improve it over time.

When starting up, the MCP Server will automatically detect a firebase.json file and the specific features used in your project and activate relevant tools.

You can optionally provide additional flags to the command:

• —dir <directory>: By default, the Firebase MCP server will look in the working directory where it’s started. Use ["--dir", "path/to/dir"] to manually configure the folder with your firebase.json file.
• —only <feature1,feature2>: The Firebase MCP Server includes many tools and will, by default, automatically activate tools for all features detected in the current project. Use e.g. ["--only", "firestore,auth"] to limit which tool categories are activated.

If you’ve never used the Firebase CLI before, you’ll need to authenticate via the CLI before you can use many of the tools integrated in the MCP server:

npx -y firebase-tools@latest login

What’s next?

The Firebase MCP server is still in its early development and we’re just getting started, future updates will enable things like:

• More tools: Additional tools to cover the whole suite of Firebase products.
• Documentation: Integration with Firebase documentation to get always up-to-date guidance.
• Deployment: Deployment and provisioning assistance for your web app, functions, etc.
• Studio: Integration with Firebase Studio.

We’re releasing the server as an experiment because we want to hear from you! Try it out with your favorite AI tools. Bugs? New tools you want to see? Let us know on GitHub!