How to connect AI assistant to SQL Server

Everyone talks about AI, all your town halls are about AI, and you’re sitting here at your desk wondering, “What do I do about AI other than rephrasing the emails and teams chats?”

I got you! Here is something exciting!
Microsoft released the Model Context Protocol (MCP) for SQL Server in June 2025.

I checked it out to see what it was all about and I was astonished. Turns out, you can connect an AI assistant like Claude directly to SQL Server and ask it questions in plain English. Claude then responds with what it finds about your connected SQL Server instance.

You will understand what I’m talking about in a few minutes!

What is MCP?

In simple English, it is a medium that lets AI interact with other tools, the tool in this case is SQL Server. So, for AI to talk to SQL Server, it needs MCP to do the interaction.

Ask questions: AI—>MCP—>SQL Server.. and sends info back .. AI<—MCP<—SQL Server.

There is MCP for all different technologies like PostGres, Kubernetes, etc, in this blog when I say MCP it is only about MCP for MSSQL server.

Alright, let’s go ahead and see how to get this all setup and working.

Setup MCP server

Install Node.js

• Use Node.js to build MCP server

• Here is the link to download Node.js - https://nodejs.org/en/download

• This will be the MCP server that connects AI to SQL Server

• There are other options to build MCP, but I’m using Node.js. Honestly, I had no idea how Node.js worked until I installed it. So if you're in the same boat, that’s totally fine, you’ll learn as you experiment, just like I did.

Install git

• To setup things on Node.js MCP server, gitbash is needed to run/build/install packages, dependencies, etc.

• Here is the link to download git - https://git-scm.com/downloads

Build MCP server

1. Clone Repo:

   I’m cloning GitHub repository from Microsoft’s blog here.
   Open gitbash and run the script below.

   /*
   https://devblogs.microsoft.com/azure-sql/introducing-mssql-mcp-server
   */
   
   git clone https://github.com/Azure-Samples/SQL-AI-samples.git

2. Go to Node project:

   After cloning, you will find Node project directory.
   On my machine, the Node project is here:

   C:\Users\harip\SQL-AI-samples\MssqlMcp\Node

   In gitbash, navigate to the appropriate path on your machine:

   cd /c/Users/harip/SQL-AI-samples/MssqlMcp/Node/

   

   3. Install Dependencies:

   Run ‘npm install’ - npm is node package manager and a library for .js packages. npm install command installs the packages and its dependencies.

   

3. After this step, a new dist folder will be created inside the Node directory:

   C:\Users\harip\SQL-AI-samples\MssqlMcp\Node\dist

4. Add Tools to the Project:

   MCP originally comes with 8 tools to connect, create, list tables etc., I added additional 14 tools from Olivar Flindall’s github. Credits to Olivar for some great ideas!!
   Download the 14 extra tools and place them in the tools directory:

   C:\Users\harip\SQL-AI-samples\MssqlMcp\Node\src\tools

5. Replace index.ts with a modified Version:

   You’ll find the original index.ts file here:

   C:\Users\harip\SQL-AI-samples\MssqlMcp\Node\src

   
   I’ve modified the following in index file:

   • Removed Entra authentication

   • Added SQL authentication using username/password

   • Enabled Trust Server Certificate

   • Added 14 tools from above

   You can download the modified index.ts here and replace the original with it.

6. Build the project:

   Run this command in Git Bash: “npm run build”
   This will rebuild index.js file in dist folder based on modified index.ts file.

   index.js file location -

   C:\Users\harip\SQL-AI-samples\MssqlMcp\Node\dist\index.js

7. Restart MCP server using - “npm run start”

   

8. MCP server is now ready to use.

Now let’s get AI tool setup.

Setup AI tool

Install Claude AI

• This is the AI assistant desktop tool to install - https://claude.ai/download

• Again you could choose either Visual Studio Copilot or Claude

• Claude config file is here - C:\Users\harip\AppData\Roaming\Claude\claude_desktop_config.json

• I modified this file to include 2 SQL Servers and gave username, password to connect with.

• Updated claude_desktop_config.json here.

• If Claude is already running, don’t just close the window, it won't fully exit.
  Terminate it from Task Manager to ensure your config changes are picked up on restart.

• Now open up claude - you should see sql servers connected and the number of tools

  

• Here are the tools that it sees, the ones from index.ts file

  

• Now you can verify by asking it to check for connectivity

  

• I caused TempDB contention(insert wicked laugh here) on SQL 2025 and asked it to check if there is any slowness on the instance. I didn’t tell Claude that I caused contention, I just asked it to checked for slowness.

  

• It does a bunch of checks

  

• Mentions about pagelatch contention in the top, as expected - SWEEEET!!!

• Ends up providing some recommendations which are not great.

  

• I did a bunch of demos including blocking, deadlocks, comparing the number of connections between instances, etc.

• My finding: It is GREAT at troubleshooting and AVERAGE at providing solutions.

• This looks like a great start to me and we could build from there.

• I’m planning to add more tools to the index for in-depth troubleshooting.

Summary

• MCP connects AI to SQL Server

• Install Node.js and Git to build and setup MCP server

• Add troubleshooting scripts/tools as needed

• Install Claude AI as an assistant tool to ask questions in plain English

• Configure Claude AI config file to connect to SQL Server instances

• Claude AI gets answers from SQL Server via MCP

• Check connectivity from Claude

• Ask questions!

Resources

• https://github.com/haripriyasb/MCPSQLServer/blob/main/src/index.ts

• https://devblogs.microsoft.com/azure-sql/introducing-mssql-mcp-server/

• https://devblogs.microsoft.com/azure-sql/mssql-mcp-server-in-action-susans-journey/

• https://oli-the-dba.com/how-i-accidentally-built-a-better-database-admin-than-myself/

• https://github.com/oflindall/MSSQLMCP_DBA/tree/main/Node

Thanks for reading,
Haripriya Naidu.