November 2025 Summaries

Firebase’s Model Context Protocol server can be installed in Google’s Antigravity IDE through the Agent window’s MCP server menu, allowing developers to connect projects to Firebase and manage related tasks through natural-language prompts. After initialization, Antigravity can create an implementation plan, request clarification when needed, and provide a walkthrough of the completed Firebase setup. The integration can support application development tasks such as retrieving web API keys, adding Firebase email-and-password authentication, and using Cloud Firestore for features like leaderboards, as demonstrated with a racing game. Antigravity can also plan and execute deployments to Firebase Hosting, including compilation and the required deployment command, without requiring users to run commands manually, while the Firebase MCP server additionally supports App Hosting for server-rendered Next.js and Angular applications.

Gemini 3 Pro preview is now accessible to Firebase Blaze-plan developers through Firebase AI Logic client SDKs for Android, Flutter, web, iOS, and Unity, allowing AI features to be built directly into client applications without server-side setup. The SDKs support capabilities including improved reasoning, function calling, encrypted thought signatures that preserve context across conversations, and higher-default-resolution multimodal input processing, with future controls planned for media resolution and thinking levels; developers can currently use thinking budgets to balance latency and cost. Firebase’s AI monitoring dashboard provides visibility into model usage, latency, error rates, request and response sizes, and traces for debugging and optimization. The integration also works with Firebase App Check to restrict unauthorized backend access and Remote Config to update models, prompts, parameters, and feature experiments without requiring app releases.

Firebase Data Connect, now renamed SQL Connect, introduces event triggers and generated admin SDKs in response to user feedback. Event triggers allow server-side code in Cloud Functions, Cloud Run, App Hosting, or other systems to run after mutations, supporting use cases such as synchronizing data with Firestore or external services, launching asynchronous tasks, and sending user notifications. Triggers can apply to every mutation or be filtered by service, connector, operation, and wildcard capture groups, although events include only mutation arguments and returned data rather than transactional pre-change snapshots. Generated admin SDKs provide type-safe, GraphQL-definition-based APIs for privileged backend and administrative environments, complementing existing arbitrary GraphQL and bulk-operation Admin SDK APIs. Because deployed SQL Connect operations are stored server-side, changes to queries or mutations require regenerating the admin SDKs and redeploying dependent services.

Crashlytics has enhanced native Android crash debugging by combining minidumps, which provide detailed symbolication for application code, with Android tombstones, which include on-device symbols for system libraries such as libc.so and libart.so. Because maintaining symbols for every system-library version across Android devices is impractical, tombstones collected by Crashlytics NDK SDK version 18.3.6 and later on Android 12 or newer supply valuable system-level context that minidumps alone cannot provide. Crashlytics now processes tombstones through its existing symbolication pipeline to restore application symbols, file names, and line numbers where possible, producing a more complete stack trace spanning both app and system frames. These paired views are available side by side in the Crashlytics dashboard and can also be exported to BigQuery, reducing the need for developers to manually retrieve or symbolicate raw tombstone data.

As of August 2026, Vertex AI is called Agent Platform, and the example demonstrates how Firebase can protect a costly AI-powered virtual try-on feature that generates outfit images using the nano banana image model. It recommends combining Firebase App Check to verify requests originate from legitimate app clients, limited-use App Check tokens that are consumed server-side to prevent replay attacks, Firebase Authentication to restrict access to authorized users, and per-user rate limiting to cap inference requests over an hour. The server accepts only a product SKU rather than user-supplied images or prompts, then retrieves the shopper profile image, product image, and predefined product prompt itself, reducing misuse and prompt-injection risks. The implementation also reuses previously generated images when available, stores outputs in Cloud Storage, validates product data, and returns user-facing errors for unauthenticated requests, consumed tokens, missing resources, quota violations, or failed image generation.

Firebase introduces Crashlytics Model Context Protocol tools that connect crash-reporting data with AI development environments such as Gemini CLI, Claude Code, and Cursor to support conversational mobile-app debugging. Crashlytics collects crash, ANR, and non-fatal error information for Android, iOS, Unity, and Flutter apps, while the MCP integration lets AI tools access this data alongside a developer’s codebase to prioritize issues, investigate stack traces and sample events, identify likely root causes, and propose or implement fixes. Using Gemini CLI requires an app with Crashlytics issues, codebase access, Node.js and npm, installation of the Gemini CLI and Firebase extension, and the `/crashlytics:connect` command to initiate a guided workflow. Developers can query filtered crash trends, such as tablet-specific issues over a selected period, ask follow-up questions about affected devices, and have Gemini analyze relevant code paths to produce detailed diagnostic reports. The approach is intended to reduce manual investigation and context switching between the Firebase console and source code while accelerating debugging and improving the end-user experience.