Can AI agents automate Firebase?

Firebase app backends are automatable today through the first-party Firebase MCP server in firebase-tools, product REST APIs, and the official Firebase CLI. This record is the Firebase identity, not Google Cloud as a whole and not Supabase or Neon. The MCP server runs over stdio as npx -y firebase-tools@latest mcp and uses the same Firebase CLI login or Application Default Credentials as the CLI. Vendor docs list write-capable tools including firebase_create_project, Firestore document CRUD, auth_get_users, auth_update_user, and messaging_send_message, plus --only feature groups. The Cloud Firestore REST API at https://firestore.googleapis.com/v1/ authenticates with a Firebase ID token or a Google OAuth 2.0 token and supports document CRUD and index administration. The Firebase Management API at https://firebase.googleapis.com uses Bearer access tokens from a service account or Application Default Credentials to add Firebase to Cloud projects and manage iOS, Android, and web apps. Firebase also documents an Auth REST API for sign-in and account management. The official Admin SDK covers Node.js, Java, Python, Go, and .NET for privileged server access. The Firebase CLI installs on Windows, macOS, and Linux, supports firebase login:ci and FIREBASE_TOKEN for non-interactive CI, and deploys Hosting, Functions, rules, and other project assets. Spark is a no-cost plan; Blaze is pay as you go with published Firestore and Functions quotas. Cloud Functions can run on HTTPS requests; vendor docs describe writing webhooks. Connector catalogs were not opened. Computer-use viability is unassessed. MCP, API, and CLI each score 8. MCP is the best path because it is the vendor agent interface with named project, Auth, and Firestore tools once CLI credentials exist.

Best path today: mcp · Overall automatability: 8/10