antigravity-mcp

name	antigravity-mcp
description	Use when Antigravity MCP is available locally and you want to query workspaces, check quota, run lightweight Antigravity asks, or generate images through the local antigravity-mcp-server.

antigravity-mcp

Use the local Antigravity MCP server for lightweight asks, workspace inspection, and image generation.

Features

• 🔌 Direct MCP Connection - Connects directly to antigravity-mcp-server
• 🖼️ Image Generation - Generate images via Antigravity AI
• 📊 Code Analysis - Analyze project code and architecture
• 🔄 Async Support - Full asyncio implementation
• 🛡️ Error Handling - Comprehensive error handling and logging

Prerequisites

# 1. Antigravity MCP Server must be installed
ls ~/.config/antigravity-mcp/bin/antigravity-mcp-server

# 2. Python 3.8+ with asyncio
python3 --version  # >= 3.8

Installation

# Skill is automatically discovered by OpenClaw
# Place in: ~/.agents/skills/antigravity-mcp/

# Verify installation
openclaw skills list | grep antigravity

Usage

Command Line

# Ping MCP server
python3 ~/.agents/skills/antigravity-mcp/scripts/antigravity_cli.py ping

# List available tools
python3 ~/.agents/skills/antigravity-mcp/scripts/antigravity_cli.py list-tools

# List known workspaces (prefers local registry)
python3 ~/.agents/skills/antigravity-mcp/scripts/antigravity_cli.py list-workspaces

# Check quota
python3 ~/.agents/skills/antigravity-mcp/scripts/antigravity_cli.py quota-status

# Ask Antigravity to generate/analyze
python3 ~/.agents/skills/antigravity-mcp/scripts/antigravity_cli.py ask \
  "生成一张太空猫咪的图片，穿着宇航服，背景是地球和星空"

# Specify output path for image tasks
python3 ~/.agents/skills/antigravity-mcp/scripts/antigravity_cli.py ask \
  "画一只可爱的猫" -o ~/Desktop/cat.png

As OpenClaw Tool

# In an OpenClaw conversation or script
from antigravity_mcp import AntigravityMCPClient

client = AntigravityMCPClient()
await client.connect()

# Generate image
result = await client.ask_antigravity(
    "生成一张架构图，展示微服务系统的数据流",
    output_path="/tmp/architecture.png"
)

await client.disconnect()

Available Actions

Action	Description	Example
ask-antigravity	Send prompt to Antigravity AI	Generate images, analyze code, answer questions
antigravity-stop	Stop running generation	Cancel long-running tasks
ping	Test connectivity	Check if server is ready
list-workspaces	List available workspaces	See available project contexts
quota-status	Check quota	Verify usage limits
launch-antigravity	Launch Antigravity	Start Antigravity with CDP

Architecture

┌─────────────────────────────────────────────┐
│         OpenClaw Agent / CLI                │
└──────────────────┬──────────────────────────┘
                   │
┌──────────────────▼──────────────────────────┐
│   antigravity-mcp Skill (Python/Asyncio)    │
│   ┌─────────────────────────────────────┐   │
│   │  AntigravityMCPClient               │   │
│   │  ├─ connect()                       │   │
│   │  ├─ ask_antigravity()               │   │
│   │  ├─ list_tools()                    │   │
│   │  └─ disconnect()                    │   │
│   └─────────────────────────────────────┘   │
└──────────────────┬──────────────────────────┘
                   │ stdio (JSON-RPC)
┌──────────────────▼──────────────────────────┐
│   antigravity-mcp-server (binary)           │
└──────────────────┬──────────────────────────┘
                   │ CDP / WebSocket / HTTP
┌──────────────────▼──────────────────────────┐
│        Antigravity AI Service               │
│   (Image Generation, Code Analysis, etc.)   │
└─────────────────────────────────────────────┘

Troubleshooting

Server not found

# Check if antigravity-mcp is installed
ls ~/.config/antigravity-mcp/bin/antigravity-mcp-server

# If not found, install Antigravity app first
# (Antigravity should install MCP server automatically)

Connection timeout

# Check if server is already running
ps aux | grep antigravity-mcp-server

# Check logs
tail ~/.config/antigravity-mcp/logs/latest.log

Image generation fails

• Verify Antigravity app has necessary permissions
• Check quota status: use quota-status action
• Ensure prompt is not blocked by safety filters

Contributing

This Skill is designed to be:

• Modular: Easy to extend with new actions
• Maintainable: Clear error handling and logging
• Compatible: Works with OpenClaw ecosystem

License

MIT - OpenClaw Agent