gemini-antigravity-bridge

Purpose

Gemini Antigravity Bridge is an MCP (Model Context Protocol) server used to establish a persistent, bidirectional communication channel between the following two ends:

• Google Gemini Spark: a cloud AI assistant with access to Google Workspace, powered by Gemini 3.8 / 3.7;
• Google DeepMind Antigravity: a local agentic IDE with full system access.

The project is optimized for Gemini 3.8's next-generation agentic reasoning capabilities, enabling Gemini Spark to dispatch coding tasks to the local machine, and Antigravity to conversely request cloud intelligence from Spark, thereby forming a fully autonomous loop.

Main Capabilities

• Connect cloud AI and the local machine via MCP using a standard, secure protocol.
• Provide 24 MCP tools, covering the following categories:
  • System execution;
  • File operations;
  • Agent orchestration;
  • Zero-trust security and telemetry (OpenAgentShield);
  • Shared memory and synchronization;
  • Spark connected apps;
  • Antigravity deep integration.
• Support autonomous file creation, terminal execution, and sub-agent orchestration.
• Integrate the OpenAgentShield zero-trust runtime sandbox, with automatic credential redaction.
• Through the bridge, Antigravity can orchestrate Spark's connected Google and third-party apps.
• Support automatic startup on Windows boot (including a silent startup script that runs automatically at login).
• Support cloud deployment for 24/7 continuous online availability.
• Support multi-layer automated monitoring via Gemini Spark scheduled tasks.

Usage

The onboarding process described in the documentation includes:

1. Prepare the environment: an ngrok account (free tier) with a reserved domain, Google Gemini Spark (Advanced/Ultra subscription), and the Google DeepMind Antigravity IDE are required.
2. Connect Gemini Spark: open Gemini Spark → Settings → Connected Apps → click "Add a custom app" → enter the public MCP URL, for example https://your-domain.ngrok-free.dev/mcp. Spark will automatically discover all 23 tools.
3. The project supports a Windows automatic startup script on boot, as well as a cloud deployment solution for round-the-clock operation.

Contribution process (according to the repository description):

Fork the repository
Create your feature branch ( git checkout -b feat/amazing-feature )
Commit your changes ( git commit -m 'feat: add amazing feature' )
Push to the branch ( git push origin feat/amazing-feature )
Open a Pull Request

The project is licensed under the Apache License 2.0.

Applicable Scenarios

• Dispatching cloud AI coding tasks to a local development machine for execution;
• A local agentic IDE conversely invoking cloud AI capabilities;
• Multi-agent workflows requiring shared memory and context synchronization across cloud and local environments;
• Automated tasks requiring orchestration of Google and third-party apps (via Spark connected apps);
• Security-sensitive agent runtime environments requiring a zero-trust runtime sandbox and credential redaction.

Limitations

• The documentation does not provide specific pricing information for free or paid tiers; apart from the license agreement (Apache 2.0), no other charging model is described.
• The documentation does not provide compatibility information for platforms other than Windows.
• The documentation does not provide specific installation steps, dependency versions, or complete code examples.
• The tool list only gives categories and does not enumerate the full names and parameters of all 24 tools one by one.
• The project references a research paper and its DOI, but the documentation does not provide performance benchmarks or effectiveness data.