One Search MCP Server to Rule Them All: Free, Clean, and Optimized for Claude Code & Cursor

If you use local AI coding tools like Claude Code CLI or Cursor, you have probably set up a web search MCP (Model Context Protocol) server at some point. And if you're like most developers, you probably ended up with three or four: one for DuckDuckGo, one for Brave Search, another for Google, and maybe a webpage reader.

Before long, you run into the classic pitfalls:

1. Context Bloat: A search tool grabs a page and dumps raw HTML, header menus, tracker codes, and styles directly into your prompt, consuming 30,000 tokens in a single request.
2. Rate Limits: You hit rate-limit errors or get blocked because you are hitting scrapers repeatedly.
3. Redundancy: You have multiple servers running when you only need one reliable, clean source of information.

Instead of writing a custom search server from scratch, you can consolidate everything into a single, clean, and free setup using the Jina AI MCP Server. Here is the configuration guide for Claude Code and Cursor.

---

Why Jina AI is the Single Best Option

Jina AI provides a dedicated Model Context Protocol server (jina-ai-mcp-server) that utilizes their search and reader endpoints (s.jina.ai and r.jina.ai).

• Zero HTML Bloat: It automatically converts search results and target webpages into clean Markdown. It strips out headers, sidebars, cookie banners, and script tags, reducing payload size by up to 90%.
• High Rate Limits (Free): While you can run it keyless (limited to 20 requests per minute), generating a free API key at jina.ai gives you 10 million free tokens and bumps your rate limit to 500 requests per minute (RPM).
• Client Guardrails: It handles automatic truncation, ensuring oversized page outputs are formatted proportionally to fit within your model's context budget.

---

Configuration Guide

Here is how to set up Jina AI as your single web search and reading tool for both Claude Code and Cursor.

1. For Claude Code CLI

Claude Code stores its global MCP configuration in your user settings file.

1. Obtain a free API key from the Jina AI Console.
2. Open your global MCP configuration file:
   • On Windows: %APPDATA%\Claude\mcp.json (or .mcp.json in your home folder)
   • On macOS/Linux: ~/.claude/mcp.json
3. Edit the file to include the Jina MCP server setup:

{
  "mcpServers": {
    "jina-ai-mcp-server": {
      "command": "npx",
      "args": ["-y", "jina-ai-mcp-server"],
      "env": {
        "JINA_API_KEY": "your_free_10M_token_api_key"
      }
    }
  }
}

2. For Cursor IDE

To configure Jina search directly within Cursor:

1. Open Cursor and navigate to Settings > Project Settings > Features > MCP.
2. Click + Add New MCP Server.
3. Configure the fields:
   • Name: jina-search
   • Type: command
   • Command: npx -y jina-ai-mcp-server
4. Under Environment Variables, add:
   • Key: JINA_API_KEY
   • Value: your_free_10M_token_api_key
5. Click Save.

Cursor will automatically start the server, exposing the search and page-reading capabilities to your Composer and Chat interfaces.

---

Alternative: Build Your Own (The DIY Blueprint)

If you still prefer a 100% custom, self-hosted Node.js MCP server that uses DuckDuckGo with no API keys, here is the blueprint of how to write it to keep the context clean:

1. Keyless Search Engine

Use a package like duck-duck-scrape to execute searches programmatically. It scrapes DuckDuckGo results cleanly without hitting key requirements or severe rate limits.

2. HTML Cleaning (Mozilla Readability)

To prevent context bloat when reading webpages, use @mozilla/readability inside your server. This parses the raw HTML and extracts only the core article text, discarding tracking banners, sidebars, and navigation elements.

3. Strict Token Truncation

Enforce character caps on the return payload:

// Truncate the page content before returning it to the LLM
const cleanText = article.textContent;
const maxChars = 8000; // Roughly 2000 tokens
const safePayload = cleanText.length > maxChars 
  ? cleanText.slice(0, maxChars) + "\n[Content truncated for context limits]"
  : cleanText;

The Verdict

Building your own server gives you total control over the cleaning logic, but it requires maintenance. For most workflows, installing the official Jina AI MCP Server and configuring it with a free API key is the fastest, cleanest, and most reliable way to achieve a single-source search environment.

Consolidate your stack, clean up your configuration files, and enjoy a bloat-free coding flow.