
Google Antigravity 2 Hands-On: Setup & First Impressions
Discover what's new in Google Antigravity 2. Read our hands-on guide to installing the app, testing AI agent tasks, and manually adding custom MCP servers
At Google I/O 2026, Google announced Antigravity 2, a standalone desktop application that treats the agent as the primary user interface. You simply describe what you want to achieve, and the agent will then generate a plan, break it down into manageable subtasks, and execute them.
Prerequisites
Before we begin, you will need:
- A computer (macOS is used in this guide)
- A modern web browser
- A Gmail account
1. Install Antigravity 2
Download Antigravity 2 from antigravity.google and install it on your computer. On the first launch, you will be prompted to sign in using your Gmail account.

Once signed in, you will notice a voice button in the input field, allowing you to record voice prompts instead of typing them.

Antigravity 2 also introduces the concept of a Project, which can span multiple local folders and operate with its own specific agent settings and permissions.


2. Set Up MCPs and Plugins
All Model Context Protocol (MCP) servers, plugins, and skills are managed under the Customizations section in the Settings menu:

Currently, the only available plugins are from Google, but hopefully, more will be released in the future.

Adding an Available MCP Server
To install a pre-configured MCP Server (e.g., GitHub):
- Click on Add MCP+.
- Select the GitHub MCP Server.



The GitHub MCP Server setup ended with an error:

Since we already have git installed on macOS, we can bypass this by executing the git command directly within the conversation. However, we first need to explicitly add the git command to our Antigravity permissions.


Let's test out the prompt to see if the git command now works:


3. Add Custom MCP Servers
If the MCP Server you want isn't on the default list, you can add it by editing the MCP configuration file directly. The configuration is located at:
~/.gemini/config/mcp_config.json
Let's add some Cloudflare MCP Servers as an example. Because Cloudflare supports dynamic client registration, we only need to enable OAuth. Add the following to your JSON configuration:
"cloudflare-api": {
"serverUrl": "https://mcp.cloudflare.com/mcp",
"oauth": {
"enabled": true
}
},
"cloudflare-docs": {
"serverUrl": "https://docs.mcp.cloudflare.com/mcp"
},
"cloudflare-bindings": {
"serverUrl": "https://bindings.mcp.cloudflare.com/mcp",
"oauth": {
"enabled": true
}
},
"cloudflare-builds": {
"serverUrl": "https://builds.mcp.cloudflare.com/mcp",
"oauth": {
"enabled": true
}
},
"cloudflare-observability": {
"serverUrl": "https://observability.mcp.cloudflare.com/mcp",
"oauth": {
"enabled": true
}
}
Go back to the Customizations section in the app and click the Refresh button. You will now see the Cloudflare MCP Servers listed, pending authorisation.

Click Authenticate next to the MCP Server. This will open an authentication page in your web browser. Once authenticated, an auth code will be generated.

Copy this auth code, paste it back into the text input field in Antigravity, and click Submit.

4. Add Custom Skills
Currently, adding custom skills cannot be done directly through the desktop application's UI. Instead, they must be added directly to the ~/.gemini/config/skills folder.
Because we use the npx skills CLI to centralise all our agent skills, any newly added skill can be made available to all agents, including Antigravity. (For a deeper dive on this, refer to our article: Solving AI Agent Skills Fragmentation with npx skills).
Initially, when typing / in the conversation box, the skills list appears empty:

It appears Antigravity 2 doesn't automatically read from the global skills folder (~/.agents/skills). To fix this, we need to create a symbolic link pointing to the Antigravity config directory:
ln -s ~/.agents/skills/excalidraw-diagram ~/.gemini/config/skills/excalidraw-diagram
Once linked, the skill becomes available in the application!

5. Test a Simple Task
Now that everything is configured, let's run a test.

As shown below, the model successfully detects and utilises the newly added Cloudflare MCP Server to handle the request.

Final Thoughts
Antigravity 2 represents a major shift from a traditional IDE to an accessible desktop application, empowering everyday users to act as orchestrators for any task.
However, Antigravity 2 needs some polish. When functionalities fail, debugging can be frustrating due to vague error handling. Additionally, the official documentation feels unrefined—likely AI-generated—as several references and file paths are currently incorrect.
Solving AI Agent Skills Fragmentation with npx skills
Running multiple AI agents like Claude Code or Kimi? Learn how to fix skills fragmentation and manage a single, centralized skills folder using npx skills CLI.
Antigravity 2 in Action: AI Agents, Skills, and MCP
Explore Antigravity 2, an agent-first desktop app. Learn to orchestrate AI subagents, use built-in skills, and leverage MCP servers to build full-stack apps.