
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.
Antigravity 2, as an agent-first desktop application, allows users to orchestrate a team of subagents to generate a daily brief from emails and calendars, or to build, test, and deploy an application. But what does this actually look like in practice?
Let’s explore what Antigravity 2 can do for you—from built-in commands and skills to how integrations like Firebase Model Context Protocol (MCP) servers are turning developers into full-stack architects. Please refer to this article, Google Antigravity 2 Hands-On: Setup & First Impressions, for setting up and configuring skills and MCP servers.
Built-in Commands
/goal
With this command, you tell the agent to run until a specific task is completely finished.
Use Case:

When we ask for dynamic subagents, three agents are spawned to complete the task.



To play the generated Textris game in HTML, please click this link:
The /goal command creates a plan, executes it step by step, and reports the progress to you at each stage. Additionally, the agent will manage the context window itself without exceeding limits. This is suitable for lengthy tasks, such as performance optimisation or module refactoring, which require multiple rounds of iteration.
/grill-me
This command triggers the skill grill-me, created by Matt Pocock. It is concise and highly effective. The idea behind it is to question you from multiple aspects, clarify edge cases, align the structures, and finalise the plan before implementation.
Use Case:



It supports multi-select capabilities:


The /grill-me command is quite impressive. It covers almost all aspects of architectural design for deploying to Cloudflare. This makes it suitable for the architectural design of an app, refactoring across multiple domains, and other potentially complex tasks.
@/
This command doesn't appear on the available list, but we sometimes need it for our context window. It is used to add a folder or files (other than media) to the conversation box.
Skills
We can type / to activate the popup box and select a skill. Let's test out excalidraw-diagram.

Use Case:



The skill was successfully executed after a couple of rounds of permission confirmations.
MCP Servers
When the agent builds a backend, you usually have to open a web console, create a database, copy API keys, and paste them back. By leveraging MCP servers directly in Antigravity, the agent gains "hands" to manipulate your cloud infrastructure.
Use Case:



The agent successfully managed to create the web app in Firebase using MCP servers. Let's check the repository.
Create a new public GitHub repository called ‘to-do-app-techvoyage’. Then push all the current project files to it with the commit message ‘Initial commit — to-do-app’.

The to-do-app-techvoyage repository is available on GitHub via this link.
Final Thoughts
While MCP and Skills in Antigravity 2—powered by Gemini 3.5 Flash—deliver blazing speed, the agentic flow is currently bottlenecked by constant permission interruptions and a lack of checkpoints for rollbacks. Until the checkpoint feature is introduced, frequent Git commits are your mandatory safety net.
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
Guide to Browser and Scheduler in Antigravity 2
Discover how the /browser and /schedule commands elevate Google Antigravity 2 into a powerful AI automation platform.