Building a Natural Language-Powered MCP SSE Server with Node.js

Posted on | In |

Recently, I built a custom MCP (Model Context Protocol) server using Node.js that supports Server-Sent Events (SSE) and an array of practical tools — including a powerful Natural Language to SQL (NL2SQL) converter. The final result was a lightweight yet extensible backend service that can be integrated with Claude Desktop, Windsurf AI, or any other MCP-compatible interface.

📌 Full source code available on GitHub

🔧 Why I Built This

I started off with some open-source examples for MCP servers (mainly focused on calculator tools), but many didn’t work out of the box. After trial and error, I landed on a minimal structure that worked and then began extending it. The most exciting addition? A tool that transforms natural language questions into executable SQL queries using Google’s Gemini API.

⚙️ What the MCP Server Can Do

Once set up, the server provides the following capabilities out of the box:

  • 📐 Mathematical Expression Evaluation: Simple calculator that evaluates expressions like 2 + 2 * 3
  • 🔁 Unit Conversion: Temperature, distance, and weight conversions between common units
  • 📆 Date Formatting: Format dates using flexible patterns
  • 🧠 Natural Language to SQL: Describe what data you want, and it returns a SQL query

🌐 Endpoints

Type URL Description
REST /mcp Main entrypoint for MCP protocol traffic
SSE /sse Server-Sent Events endpoint (MCP over SSE)

🧠 Natural Language to SQL with Gemini

One of the key tools in this project is nl-to-sql. You describe your query in plain English, and it generates valid SQL based on pre-loaded schemas.

🧪 Example:

const result = await nlToSql({
  query: "Find all employees in the IT department with salary greater than 70000"
});

Under the hood:

  1. Server loads all table schemas from /src/schemas/*.sql
  2. Your query and schema info are packed into a prompt
  3. The prompt is sent to Gemini API
  4. The resulting SQL is parsed and returned

✅ Supported Tables

  • employees
  • departments
  • projects
  • employee_projects

This tool is especially handy for building AI interfaces that connect natural language to real data.

🖥️ Claude Desktop Integration

You can connect this server to Claude Desktop by editing your configuration like this:

{
  "mcpServers": {
    "RustamMCP-Server": {
      "command": "npx",
      "args": ["mcp-remote", "http://localhost:3000/mcp"]
    }
  }
}

After restarting Claude, your tools (like calculate, convert, nl-to-sql) will show up.

✅ Tested with Windsurf AI, which recognized and used the tools successfully

🛠️ Local Setup

Clone the repo and run:

npm install
cp .env.example .env
# Add your GEMINI_API_KEY in the .env file
npm run dev

Server starts on http://localhost:3000.

💡 What’s Next?

Some next steps or ideas:

  • ✅ More schemas and domain-specific NL2SQL patterns
  • 🔍 Logging and analytics of tool usage
  • 🔐 Rate limiting and authentication for hosted setups
  • 🧩 Adding custom AI model backends (e.g. Claude or local LLMs)

📂 Final Thoughts

Building this project helped me appreciate the modularity of MCP and the power of Gemini’s generative models. What started as a calculator became an AI-powered backend for natural language interfaces.

Feel free to fork the GitHub repo, experiment with your own tools, or deploy a custom version.

Stay tuned for more features and real-world integrations.

🚀 Built with Node.js, Express, Gemini API, and the MCP SDK