DEV Community: z z

I built an MCP server that searches the web

z z — Sun, 24 May 2026 09:47:32 +0000

MCP servers become a lot more useful when they can reach the internet. A file-reader is nice. A tool that fetches live data is something else.

I built a web search tool for MCP. It uses DuckDuckGo (free, no API key needed) and returns LLM-formatted results.

The tool

The tool accepts a query and optional maxResults. It searches DuckDuckGo's HTML endpoint (no API key required) and returns titles, snippets, and URLs.

The code pattern

ypescript async function searchDuckDuckGo(query: string, maxResults = 5) { const url = https://html.duckduckgo.com/html/?q=; const res = await fetch(url); const html = await res.text(); // parse results... }

DuckDuckGo's HTML endpoint works without API keys. No rate limiting for reasonable use.

Adding it to your server

The full implementation with error handling, timeout, and result formatting is in the premium MCP kit. The free starter has the basic pattern.

Why this matters

An MCP server that stays local is a typewriter. An MCP server that reaches the web is a computer. Your LLM can check current prices, look up documentation, verify facts -- while you watch.

The starter repo is here.

Connect your MCP server to Claude Desktop in 5 minutes

z z — Sun, 24 May 2026 09:46:48 +0000

You built an MCP server. Now what? Here's how to wire it up to Claude Desktop so you can actually use those tools.

Step 1: Build it

ash git clone https://github.com/zhirenhun-stack/mcp-server-starter.git cd mcp-server-starter npm install && npm run build

Step 2: Find Claude Desktop config

On Mac:
json ~/Library/Application Support/Claude/claude_desktop_config.json

On Windows:
json %APPDATA%\Claude\claude_desktop_config.json

Step 3: Add your server

json { "mcpServers": { "my-starter": { "command": "node", "args": ["/path/to/mcp-server-starter/dist/index.js"] } } }

Restart Claude Desktop. You should see a hammer icon in the bottom right.

What you can do now

Once connected, Claude can:

Read files from your machine
List directories
Search for code patterns

All through natural language. Ask "find all TODO comments in my project" and it runs grep, reads the files, and summarizes.

Add your own tools

The starter has three tools. You'll want more. The pattern is always:

ypescript server.setRequestHandler(CallToolRequestSchema, async (request) => { const { name, arguments: args } = request.params; // your code here });

Each tool needs: a name, an input schema (JSON Schema describing what arguments it expects), and a handler function. That's it.

From starter to production

The free starter gives you the basics. If you're building something for clients or production, the premium version adds:

API integration with timeout handling
Web search via DuckDuckGo
Dockerfile for deployment
One-command VPS deploy script
Commercial license

Where MCP is going

Every major LLM client is adding MCP support. Claude Desktop has it. Cursor is adding it. VS Code has it via extensions. Write your tools once, and they work everywhere.

The protocol is simple enough that the starter is about 80 lines of TypeScript. You can read the whole thing in 10 minutes.

I built a minimal MCP server starter so you don't have to

z z — Sun, 24 May 2026 09:09:06 +0000

MCP (Model Context Protocol) is what Anthropic built so LLMs can talk to your tools. Write your server once, connect it to Claude Desktop, Cursor, VS Code -- anything that speaks MCP.

I got tired of copying the same boilerplate, so I stripped it down to the bare minimum.

What you get

npm install && npm run build && npm start

Three tools out of the box:

read_file - read any text file
list_directory - list files with sizes
grep_search - search for text across your codebase

TypeScript, official MCP SDK, stdio transport. No HTTP server, no config files.

The pattern

Every MCP tool follows the same shape: a name, an input schema (JSON Schema), and a handler function. The SDK handles the protocol plumbing.

Open source

The starter is on GitHub. MIT licensed.

I also sell a premium version with API integration, web search, Dockerfile, and one-command deploy. If that saves you an afternoon: Gumroad