Web Search MCP Server

A Web Search MCP Server is a lightweight, TypeScript‑based MCP (Model Context Protocol) server that gives locally‑hosted LLMs real‑time, on‑premise web‑search capabilities. It eliminates the need for paid APIs by directly connecting to popular search engines—Bing, Brave, and DuckDuckGo—and extracting full page content or lightweight snippets.

Why Use It?

Quick Installation

# 1. Clone the repo
git clone https://github.com/mrkrsl/web-search-mcp.git
cd web-search-mcp

# 2. Install Node.js v18+ and npm v8+
#    (Use official installers or nvm to manage versions.)

# 3. Install dependencies & Playwright browsers
npm install
npx playwright install

# 4. Build the TypeScript code
npm run build

# 5. Run the server (development mode)
npm start   # or "npm run dev" for hot‑reload

Tip: On Windows, replace forward slashes and use PowerShell commands.

Configuration

Create or edit mcp.json in your project root (or any location you prefer) to tell your MCP client where to find the server binary.

{
  "mcpServers": {
    "web-search": {
      "command": "node",
      "args": ["/path/to/web-search-mcp/dist/index.js"],
      "env": {
        "MAX_CONTENT_LENGTH": "10000",
        "BROWSER_HEADLESS": "true",
        "MAX_BROWSERS": "3",
        "BROWSER_FALLBACK_THRESHOLD": "3"
      }
    }
  }
}

• command – typically node.
• args – absolute path to dist/index.js.
• env – optional environment variables for fine‑tuning.

Common Environment Variables

MCP Tool Overview

The server exposes three distinct tools that can be called through any MCP‑compatible client (LM Studio, LibreChat, or your own application).

1. full-web-search

Example request

{
  "name": "full-web-search",
  "arguments": {
    "query": "TypeScript MCP server",
    "limit": 3,
    "includeContent": true
  }
}

2. get-web-search-summaries

Same search strategy as full-web-search but returns only the snippet/descriptions—no full page content. Perfect for quick lookup or when bandwidth is limited.

Example request

{
  "name": "get-web-search-summaries",
  "arguments": {
    "query": "best TypeScript web search",
    "limit": 5
  }
}

3. get-single-web-page-content

Useful when you already know the desired URL and just need the main content.

Example request

{
  "name": "get-single-web-page-content",
  "arguments": {
    "url": "https://example.com/article.html",
    "maxContentLength": 5000
  }
}

Using with a Local LLM Client

Below is a minimalist example for integrating the server into a LibreChat configuration.

# librechat.yaml
mcpServers:
  web-search:
    type: stdio
    command: node
    args:
      - /app/mcp/web-search-mcp/dist/index.js
    env:
      MAX_CONTENT_LENGTH: "10000"
    serverInstructions: true

In a chat message:

{{{ full-web-search(query="nodejs web search", limit=5, includeContent=true) }}}

The LLM will receive the structured JSON with URLs, titles, descriptions, and full content, which can be fed as context for generation.

Performance Tips

Remember: Tuning these values depends on your network, CPU, and the size of LLM requests.

Troubleshooting

Extending the Server

The codebase is modular. New tools can be added easily:

1. Create a file under src/tools/ implementing a run function.
2. Export it in src/index.ts.
3. Update the package.json and mcp.json if needed.

Tip: Leverage the zod schema validation already present to enforce arguments.

Licensing

The project is licensed under the MIT License. Feel free to fork, modify, and incorporate into your own pipelines.

Final Words

The Web Search MCP Server closes a significant gap for local LLM developers: instant, API‑free web browsing. By combining multiple search engines, sophisticated content extraction, and a clean MCP interface, it offers both depth (full page content) and speed (summary mode) in a single repository. Whether you run a small research assistant or a production chatbot, this tool gives your LLM the up‑to‑date knowledge it needs without external costs.

Happy searching!