Google just announced Gemini 3 Pro and the Antigravity release. I am actually more excited about Antigravity because of Varun Mohan. I have been a big fan of Windsurf, and I had a feeling from the start that once Varun joined the team, something amazing would arrive.
And here we are with Antigravity.
I have been experimenting with it for the past few hours and here are my thoughts:
What is Antigravity?
Google Antigravity is a new Agentic development platform from Google that changes how software is built by letting developers work at a higher task-focused level. With the reasoning and Agentic coding abilities of Gemini 3, Antigravity allows AI Agents to plan, code, and validate complete software tasks on their own.
It includes an Agent manager dashboard, a VS Code-style editor, and deep browser connectivity, letting Agents interact with web apps directly for testing and validation. Instead of being just another coding tool, it becomes a true development partner that improves productivity by managing multiple projects and tasks in parallel with autonomous workflows.
Download now: https://antigravity.google/download
Important notes to improve your Vibe Coding with Antigravity
If you want to start with Antigravity, first install it and pick your preferred IDE layout. After setup, move to planning before writing any code.
- Begin by defining your task and the features you need. Then prepare an implementation plan and workflow. If you already have clarity, write it yourself. If not, simply describe the goal and let Gemini 3 generate the Implementation Plan, Task, and Walkthrough files.
- IMPORTANT: DO review the Implementation plan, Task list, and Walkthrough. If there is any task related to testing, remove that. Do tell the Agent not to run tests automatically.
- Do the testing manually because automated testing from the Agent can consume more tokens than required.
- When Vibe coding, the key stage is implementing features. Avoid asking the Agent to build all features in a single run. If you have more than 5 features, begin with the first one and then move step by step.
- After completing the first session, ask Gemini to generate a SUMMARY.md of the conversation. Save this summary since you can reuse it for future sessions without repeating the entire process.
- Once you have two features ready and you have already reviewed the Implementation plan and workflow, start a new task using the saved SUMMARY as context and follow the same process.
- Now for further improvements, add Custom MCP servers that keep you updated with documentation and SDK references for your tech stack. MCP Servers can help fetch context for better code generation and the latest.
Lets see how to add your own custom MCP Server in Google Antigravity.
How to add a Custom MCP server in Google Antigravity
- Click on Agent session and select the “…” dropdown at the top of the editor’s side panel. And select MCP Servers, you will find the MCP Store.
2. To add a custom MCP server, select Manage MCP Servers at the top of the MCP store and then click on View raw config in the main tab.
3. Modify the mcp_config.json with your custom MCP server configuration. Note for Vibe coding, having these 3 MCP servers can be handy:
- Context-7: Up-to-date Code Docs for the tech stack your app needs.
- Sequential Thinking: For dynamic and reflective problem-solving through a structured thinking process.
- Retriever-based MCP: Save and retrieve the code that works for future purposes. for example: Qdrant MCP Server.
NOTE: Sequential Thinking is already in MCP Store, just click on Install.
mcp_config.json:
Syntax for npx and uvx based MCP
{
"mcpServers": {
"sequential-thinking": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-sequential-thinking"
]
},
"context7": {
"command": "npx",
"args": [
"-y",
"@upstash/context7-mcp",
"--api-key",
""
]
},
"qdrant": {
"command": "https://medium.com/Users/tarunjain/.local/bin/uvx",
"args": [
"https://medium.com/Users/tarunjain/Desktop/mcp-server-qdrant"
],
"env": {
"QDRANT_URL": "",
"QDRANT_API_KEY": "",
"COLLECTION_NAME": "vibe-coding",
"EMBEDDING_MODEL": "sentence-transformers/all-MiniLM-L6-v2"
}
}
}
}
Get your Qdrant URL and API key after creating a free cluster on: Qdrant Cloud
Get your Context-7 API key from here: Context 7 Dashboard
4. Now, come back to Managed MCP servers and click Refresh.
We have 5tools across 3 MCP Servers. Context-7 have 2 tools, Qdrant have 2 tools and Sequential Thinking have one tool.
If you are new to MCP, watch my video where I use the Qdrant MCP server with Claude Desktop and explain what MCP is, why you need it, and how it works.
https://medium.com/media/486193a4804c2c62f1bf0d6a7d624a33/href
Demo
Prompt
Plan: To build a 2 player game, but the catch is, the 2 players here are the LLM.
Scene-1: Chose 2 players. we need to have options selectbox to pick the LLM i.e., Gemini, OpenAI or Claude as of now.
Each LLM choice needs to have to options: Enter API key and Choose the model name.
Scene-2: Now I need to have 2 games: Tic-Tac-Toe and Reversi: Both 2 player games. Start with Tic-Tac-Toe first.
Scene-3: Once the game is selected, in scene 3 I need the visuals, to check what move was made by what, then end the results and show the winner. Make sure to maintain the context well for each LLM.
For LLM inference use LiteLLM:
import os
from litellm import completion
os.environ[”OPENAI_API_KEY”] = “your-api-key”
response = completion(
model = “gpt-4o”,
messages=[{ “content”: “Hello, how are you?”,”role”: “user”}]
)
os.environ[’GEMINI_API_KEY’] = “”
response = completion(
model=”gemini/gemini-pro”,
messages=[{”role”: “user”, “content”: “write code for saying hi from LiteLLM”}]
)
os.environ[”ANTHROPIC_API_KEY”] = “your-api-key”
messages = [{”role”: “user”, “content”: “Hey! how’s it going?”}]
response = completion(model=”claude-opus-4-20250514”, messages=messages)
print(response)
First start with Tic-Tac-Toe and then Reversi. One game at a time.
App Screenshot
🔗 GitHub Repo — LLM Game Vibe Coded: https://github.com/lucifertrj/llm-game-battle
VIDEO TUTORIAL COMING SOON. Subscribe to YouTube
https://youtube.com/@aiwithtarun
Google Antigravity: How to add custom MCP server to improve Vibe Coding was originally published in Google Developer Experts on Medium, where people are continuing the conversation by highlighting and responding to this story.
