Agentic Coding in Astro Projects Using Antigravity

Introduction to Agentic Coding

The landscape of AI-assisted development is shifting from autocomplete tools to agentic coding. Instead of writing code snippet-by-snippet, developers can now describe desired features or debugging targets, and let an autonomous agent carry out the tasks in a loop: reading files, executing commands, checking for linting issues, and iterating until the goal is met.

Google’s Antigravity CLI (agy) is a terminal-based interface designed to bring this agentic power to your projects. In this post, we’ll explore how to set up and optimize Antigravity for an Astro web application.

1. Setting Up Antigravity in an Astro Project

To configure Antigravity for your workspace, the CLI expects a configuration structure. You can customize the agent’s behavior globally or locally.

Workspace Configuration

By following the Antigravity workspace rules, you can place a .agents folder at the root of your Astro repository. Inside this folder, you can store:

• instructions.md: Rules and standards that the agent must adhere to (e.g., using UnoCSS, preserving documentation integrity).
• mcp_config.json: Local Model Context Protocol server endpoints specific to this project.

For example, your /.agents/instructions.md might look like this:

# Local Agent Guidelines
- Astro Version: 5.x
- Content Management: Use the new Content Layer API (`src/content.config.ts`).
- Styles: Always use UnoCSS utility classes.

2. Leveraging the Astro Docs MCP Server

One of the key features of the Antigravity ecosystem is the Model Context Protocol (MCP). Instead of relying on outdated training data, your agent can connect to the official Astro Docs MCP Server to query the latest documentation in real time.

You can configure the server URL remotely by pointing your agent to: https://mcp.docs.astro.build/mcp

Once connected, your agent will always use up-to-date Astro 5.x standards, such as:

• Using the render() helper from astro:content instead of the deprecated .render() entry method.
• Constructing Content Layer schemas using the glob loader.
• Configuring internationalization path overrides.

3. Best Practices for Agentic Development in Astro

When pair-programming with an agent in an Astro codebase, keep these rules in mind:

1. Keep the Rules Local: Storing instructions in the /.agents directory of your workspace root ensures they are checked into Git and shared with other developers.
2. Explicit Intent: When triggering the agent, clearly specify the boundaries of your files. Pointing the agent to files via absolute or relative links helps it find files faster.
3. Run Checks Continuously: Use the agent to run astro check and eslint commands synchronously to catch syntax and TypeScript issues before committing your work.

Agentic coding with Antigravity turns development from a manual coding cycle into a high-level system architectural guide, letting you focus on design and user experience while the agent implements the clean code.