SERP MCP: Local Google Search for AI Agents

TL;DR

SERP MCP is an open source MCP server that scrapes Google search results with automatic fingerprint rotation. It runs locally on your machine, giving you direct access to Google SERPs for Claude Code or any MCP-compatible client.

Links:

• GitHub Repository
• PyPI Package

What is SERP MCP?

SERP MCP is a Model Context Protocol (MCP) server that lets AI agents search Google directly. Unlike API-based solutions, it scrapes Google's search results page locally using browser fingerprint rotation to avoid detection.

Key Features

• Fingerprint Rotation - Automatic browser fingerprint rotation per request
• Lite Mode - 50% less traffic by extracting only organic results (default)
• Full Mode - Extracts organic results, sitelinks, PAA, related searches, knowledge graph
• Location Support - Encode locations using protobuf (UULE format)
• Country & Language - 243 country codes and 78 language codes supported
• Time Filtering - Filter by hour, day, week, month, or year
• MCP Protocol - Works with Claude Code, opencode, and any MCP client

Installation

Claude Code

claude mcp add serp-mcp -- uvx serp-mcp

opencode

Add to your ~/.config/opencode/opencode.jsonc:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "serp-mcp": {
      "type": "local",
      "command": ["uvx", "serp-mcp"],
      "enabled": true
    }
  }
}

Manual Usage

# Run with uvx (automatically installs latest version)
uvx serp-mcp

Development

git clone https://github.com/LiranYoffe/serp-mcp
cd serp-mcp
uv sync
uv run serp-mcp

Usage

The MCP server exposes a search tool with these parameters:

Quercle vs SERP MCP: When to Use Each

Both tools provide search capabilities for AI agents, but they solve different problems:

When to Use Quercle

• You're building production agents - Reliable API with SLA, no infrastructure to manage
• You need answers, not just URLs - Quercle's search tool visits pages and synthesizes information with citations
• Security matters - Prompt injection detection protects your agent
• You need page content - The fetch tool extracts and analyzes any webpage
• You want both tools - Search for synthesized answers, fetch for reading specific pages

When to Use SERP MCP

• You're developing locally - Free, no API key needed, great for experimentation
• You need raw SERP data - URLs, snippets, featured snippets, PAA boxes, knowledge panels
• You want to process results yourself - Feed raw data to your own LLM or pipeline
• You need specific Google features - Sitelinks, related searches, knowledge graph
• You're behind a residential proxy - Reduces detection risk for higher volume

Deployment Options

Personal Use

Run locally with uvx serp-mcp for personal Claude Code sessions.

Behind a Residential Proxy

For higher volume or reduced detection risk:

HTTP_PROXY=http://user:pass@proxy:port uvx serp-mcp

As an API Service

SERP MCP can be wrapped as an HTTP API for team use, though you'll need to manage proxies and rate limiting.

Contributing

SERP MCP is open source and contributions are welcome:

github.com/LiranYoffe/serp-mcp