anti-search

MCP server for web search and URL reading via CLIProxyAPI Antigravity backend.

Features

• web_search - Google Search via Gemini's googleSearch tool
• read_url - Read and summarize content from any URL (including Reddit, news sites, etc.)
• grounded_search - Combined search + URL context for comprehensive research

Prerequisites

• CLIProxyAPI running with Antigravity login
• Bun runtime

Installation

# Via npx (recommended)
npx anti-search

# Or via bunx
bunx anti-search

# Or install globally
npm install -g anti-search

Configuration

Set environment variables:

# CLIProxyAPI endpoint (default: http://localhost:8317)
ANTI_BASE_URL=http://localhost:8317

# API key for CLIProxyAPI (default: dummy)
ANTI_API_KEY=dummy

# Model to use (default: gemini-2.5-flash)
ANTI_MODEL=gemini-2.5-flash

Usage with Claude Code

Add to your MCP configuration (~/.claude.json or lazy-mcp config):

{
  "mcpServers": {
    "anti-search": {
      "command": "npx",
      "args": ["anti-search"],
      "env": {
        "ANTI_BASE_URL": "http://localhost:8317",
        "ANTI_API_KEY": "dummy"
      }
    }
  }
}

Tools

web_search

Search the web using Google Search.

{
  "name": "web_search",
  "arguments": {
    "query": "latest AI news 2025"
  }
}

read_url

Read and summarize content from a URL.

{
  "name": "read_url",
  "arguments": {
    "url": "https://www.reddit.com/r/LocalLLaMA/",
    "instruction": "List the top 3 posts"
  }
}

grounded_search

Search the web and read URLs for comprehensive research.

{
  "name": "grounded_search",
  "arguments": {
    "query": "Claude Code features and capabilities"
  }
}

How It Works

This MCP server connects to CLIProxyAPI which provides access to Gemini's:

• googleSearch - Server-side Google Search integration
• urlContext - Server-side URL content retrieval

Both features are part of Google's Gemini API grounding capabilities.

License

MIT