MCP 服务器

serp-it

An MCP server that aggregates web search results from multiple engines and optionally renders pages to Markdown, providing a unified search interface.

README

serp-it

serp-it is an MCP (Model Context Protocol) server that exposes web search and optional page rendering capabilities through a simple stdio interface.

Features

🔍 Multi-Engine Search: Aggregates results from 7 search engines (AOL, Brave, Bing, DuckDuckGo, Yahoo, Startpage, Yandex)
📄 PDF Support: Automatically detects and extracts text from PDF documents
🌐 Page Rendering: Converts web pages to Markdown using headless Chromium
♻️ Browser Pooling: Efficient browser instance management for better performance
🗺️ Region Support: Search with locale/region codes (e.g., en-US, en-GB)
🔄 Result Deduplication: Merges duplicate results across engines
📦 Docker Ready: Includes Dockerfile for deployment with supergateway

AOL Search Integration

The AOL search engine often flies under the radar but returns unique SERP blends that combine Yahoo ranking signals with syndicated content. serp-it includes first-class AOL support with the following behavior:

Requests are issued against https://search.aol.com/aol/search with JavaScript disabled (nojs=1) so that results are stable for scraping.
Region hints are translated into Accept-Language headers (en_US -> en-US), giving you localized snippets where AOL supports them.
Result URLs are unwrapped with the same redirect decoder used for Yahoo, ensuring clean, direct links instead of AOL's tracking intermediaries.
Duplicate links within the same result set are filtered before they reach the aggregator, protecting downstream clients from redundant entries.

If AOL should be temporarily disabled (for example during rate limiting) you can comment it out in src/SearchAggregator.ts and the rest of the stack will continue operating.

Supported Engines

Engine	File	Notes
AOL	`src/engines/Aol.ts`	HTML response, Yahoo-style redirect cleanup.
Brave	`src/engines/Brave.ts`	Rapid updates, privacy-focused snippets.
Bing	`src/engines/Bing.ts`	Provides richer entity cards.
DuckDuckGo	`src/engines/DuckDuckGo.ts`	No personalization, global focus.
Yahoo	`src/engines/Yahoo.ts`	Often mirrors AOL but with different ordering.
Startpage	`src/engines/Startpage.ts`	Google-backed privacy results.
Yandex	`src/engines/Yandex.ts`	Strong non-English indexing.

Installation

npm install
npm run build

Playwright Setup

# Install Chromium used by Playwright
npm run playwright:install

# (Linux/WSL) install required system libraries
npm run playwright:install-deps

Usage

Run as MCP Server (stdio)

npm start

The server exchanges JSON-RPC 2.0 messages over stdio.

Tools

search: Aggregate results from the configured engines.
search_fetch: Search and render the top matches to Markdown.

Inputs:

query (required string): Search query text.
region (optional string): Locale code such as en-US or fr-FR.
maxFetch (optional number, search_fetch only): Number of rendered pages (default 5, max 10).

Outputs:

Array of enriched search results (title, URL, snippet, engine metadata).
Separate error lists for engine failures and fetch/render issues.

Development

# Rebuild on change
npm run dev

# Remove build artefacts
npm run clean

# Full clean build
npm run rebuild

# Run the integration tests
npm test

Docker Deployment

With supergateway

docker build -t serp-it-mcp .
docker run -p 8080:8080 serp-it-mcp

The container image bundles the MCP server alongside supergateway to expose SSE/HTTP endpoints on port 8080.

Environment Variables

PLAYWRIGHT_CHROMIUM_EXECUTABLE_PATH: Override the Chromium binary path.
PLAYWRIGHT_SKIP_BROWSER_DOWNLOAD: Set to 1 to reuse a preinstalled browser.

Configuration

Claude Desktop (direct stdio)

Add the server to claude_desktop_config.json when running it locally via node:

{
  "mcpServers": {
    "serp-it": {
      "command": "node",
      "args": ["/path/to/serp-it/dist/index.js"]
    }
  }
}

Claude Desktop (Docker + supergateway SSE)

When you run the provided container (or start-gateway.sh), supergateway exposes an SSE bridge at http://localhost:8080/sse. Point Claude Desktop at that endpoint by updating claude_desktop_config.json:

{
  "mcpServers": {
    "serp-it-docker": {
      "type": "sse",
      "url": "http://localhost:8080/sse"
    }
  }
}

Adjust the host/port if you map the container differently. You can also add a headers object inside the block if your gateway requires authentication.

Claude Desktop (spawn with npx)

You can let Claude Desktop launch the server through npx, which downloads (or reuses) the published package on demand:

{
  "mcpServers": {
    "serp-it": {
      "command": "npx",
      "args": ["-y", "serp-it"]
    }
  }
}

If you are working from a clone of this repository, run npm run build first so that the serp-it binary points at dist/index.js.

Generic MCP Client (TypeScript)

import { Client } from '@modelcontextprotocol/sdk/client/index.js';
import { StdioClientTransport } from '@modelcontextprotocol/sdk/client/stdio.js';

const transport = new StdioClientTransport({
  command: 'node',
  args: ['./dist/index.js']
});

const client = new Client({
  name: 'my-client',
  version: '1.0.0'
}, {
  capabilities: {}
});

await client.connect(transport);