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.
Top comments (0)