Antigravity MCP Server
Integrates with Google Antigravity/Gemini via the locally running, signed-in Antigravity app. The MCP server sends prompts to Antigravity and returns Gemini model responses, with support for model selection and multi-turn conversations through a session id.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@Antigravity MCP Serverreview this Python function and suggest performance improvements"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Antigravity MCP Server
Bridge a local Google Antigravity subscription to MCP coding agents — Cursor, Claude Code, Codex, and any other MCP client. No extra API bill. The call stays on your machine, on the session Antigravity already authenticated.
Antigravity must be installed, signed in, and running before the server starts. This project is not affiliated with Google.
What you get
Existing subscription. The server talks to the local Antigravity language server. It does not mint a separate API key.
Multi-turn memory. Each reply ends with a session id. Pass it back as
conversation_id.Full replies. Long answers are read from
transcript_full.jsonl, not the truncated compact log.Remote hosts. An SSH
RemoteForwardlets a Linux dev machine reach the Mac server. Nothing extra is installed on the remote side.
Related MCP server: supergravity-mcp
How a request moves
Every call takes the same last two steps. Only the first step changes, depending on where the agent is running.
The agent sends your prompt to
server.py.server.pyhands it to the Antigravity app already open on your Mac.Antigravity uses that signed-in session to reach Gemini, and the answer comes back.
Agent on the same Mac. Cursor, Claude Code, or Codex starts server.py itself. Nothing listens on a port.
Agent on a remote machine. Start server.py --sse on the Mac first. It listens at http://127.0.0.1:8999/sse. An SSH RemoteForward makes that same address work on the remote machine. The remote agent never talks to Gemini directly.
flowchart TB
mac["Cursor, Claude Code, or Codex on the Mac"] --> server["server.py"]
remote["Agent on a remote machine"] --> tunnel["SSH forwards port 8999"]
tunnel --> server
server --> app["Antigravity app on the Mac"]
app --> gemini["Gemini"]Install
Requires macOS, Python 3.10+, and the Antigravity app open.
git clone https://github.com/qiansexingqiu/antigravity-mcp.git
cd antigravity-mcp
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txtPrint the two absolute paths the configs below expect. ~ is not expanded in most MCP command fields, so paste these values as-is.
echo "python $(pwd)/.venv/bin/python"
echo "server $(pwd)/server.py"Connect a client
Replace both /ABS/PATH values with the paths printed above. Start Antigravity first, then reload the client.
Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"antigravity": {
"command": "/ABS/PATH/.venv/bin/python",
"args": ["/ABS/PATH/server.py"]
}
}
}Reload the window: Command Palette → Developer: Reload Window. The tool name is ask_antigravity.
Claude Code
User scope, available in every project:
claude mcp add --scope user --transport stdio antigravity -- \
/ABS/PATH/.venv/bin/python \
/ABS/PATH/server.pyThe same entry, if you edit .mcp.json (this project) or ~/.claude.json (all projects) by hand:
{
"mcpServers": {
"antigravity": {
"type": "stdio",
"command": "/ABS/PATH/.venv/bin/python",
"args": ["/ABS/PATH/server.py"]
}
}
}Codex
codex mcp add antigravity -- \
/ABS/PATH/.venv/bin/python \
/ABS/PATH/server.pyOr add this to ~/.codex/config.toml:
[mcp_servers.antigravity]
command = "/ABS/PATH/.venv/bin/python"
args = ["/ABS/PATH/server.py"]Remote SSH
Use this when the agent runs on another machine and Antigravity stays on the Mac. Codex's HTTP MCP setting is streamable HTTP, not this server's SSE endpoint, so the remote example below is for Cursor and Claude Code.
1. On the Mac, leave Antigravity open and start the SSE server. It listens only on loopback: http://127.0.0.1:8999/sse.
cd /ABS/PATH
.venv/bin/python server.py --sse --port=89992. On the Mac, add a reverse forward to ~/.ssh/config, then reconnect:
Host dev-server
HostName 203.0.113.10
User ubuntu
RemoteForward 8999 127.0.0.1:89993. On the remote machine, point the client at the forwarded port.
Cursor, ~/.cursor/mcp.json:
{
"mcpServers": {
"antigravity": {
"url": "http://127.0.0.1:8999/sse"
}
}
}Claude Code, .mcp.json or ~/.claude.json:
{
"mcpServers": {
"antigravity": {
"type": "sse",
"url": "http://127.0.0.1:8999/sse"
}
}
}Tool: ask_antigravity
Parameter | Type | Default | Meaning |
| string | required | The question or code to send. |
|
|
|
|
| string | omitted | Session id from an earlier reply. Omit it to start a new conversation. |
Ask for it by name. A first turn:
Use ask_antigravity with model flash.
Design a high-throughput micro-batching buffer in Python asyncio
with graceful shutdown.The reply ends with a session id. A follow-up in the same conversation:
Use ask_antigravity.
conversation_id: 13278f76-619d-48dd-bf7a-14114c41146c
How should this buffer apply backpressure if the downstream flush fails?
Keep the answer to a few lines of code.License
This server cannot be deployed
Maintenance
Related MCP Connectors
- QuallaaOAuthcom.quallaa
Talk to your public-facing AI from any MCP client — Claude, ChatGPT, Cursor, Cline, Windsurf.
A comprehensive Model Context Protocol (MCP) server that enables AI assistants to interact with yo…
Remote streamable-HTTP MCP server running on a single Cloudflare Worker. Your assistant gets live Airbnb, Amazon, Booking.com, Google Flights, Maps and Reddit data, social search on X, Instagram and TikTok, the Meta Ad Library, and image/video generation without any keys. Connect your own accounts to let it send WhatsApp or Telegram messages, work an IMAP inbox, manage Meta Ads campaigns and publish to X and LinkedIn. OAuth 2.1 with PKCE; stored credentials are AES-256-GCM encrypted.
Enable secure connectivity between Sentry issues and debugging data, and LLM clients, using a Model Context Protocol (MCP) server.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceAntigravity Link is an MCP server and mobile companion for Google's Antigravity IDE. It captures live snapshots of the Antigravity chat UI via Chrome DevTools Protocol and serves them to a mobile web client. From your phone you can send messages, upload files, stop AI generation, switch instances, and read task/walkthrough/plan documents. Exposes 9 MCP tools and a full OpenAPI 3.1 spec.25 npm213MIT
- AlicenseAqualityDmaintenanceAn MCP server that lets Claude Code delegate tasks to the Google Antigravity desktop app by driving its real UI, providing a bridge for Intel Macs without the official CLI.212 npm1MIT
- FlicenseNot gradedqualityCmaintenanceEnables local AI agents to generate code and interact with Google Antigravity (Gemini Pro) via MCP, consuming zero API tokens.-
- AlicenseNot gradedqualityCmaintenanceThis MCP server bridges Codex to Antigravity CLI, allowing Codex to delegate coding tasks (e.g., code modification, testing, and fixing) to Antigravity within specified project directories.MIT