From zero to AI-powered database queries in 10 minutes
Most database tutorials start with SQL. This one starts with a question.
"How many devices haven't checked in for more than 30 days?"
That's the kind of question your IT manager, your CTO, or your most annoying PM asks on a Tuesday afternoon. Normally, it sends someone to a terminal. It takes 10 minutes to write the query, 2 minutes to run it, and 5 minutes to explain the output.
With Conexor + an MCP-compatible AI like Claude, that's now a 10-second conversation.
Here's how to set it up.
What you need
- A PostgreSQL or MySQL database
- A Claude account (or any MCP-compatible AI client)
- A Conexor account — free tier available
Step 1 — Connect your database to Conexor
Head to conexor.io and create a new connection. You'll need your database host, port, username, and password.
Conexor runs an on-premise agent alongside your database — which means your credentials and data never leave your infrastructure. The agent handles the MCP protocol layer. You stay in control.
Connection setup takes about 3 minutes.
Step 2 — Get your MCP endpoint
Once connected, Conexor generates a unique MCP endpoint for your database. Copy it. This is what your AI client will use to reach your data.
Step 3 — Configure Claude Desktop (or your preferred MCP client)
Open your Claude Desktop config file:
~/Library/Application Support/Claude/claude_desktop_config.json
Add your Conexor endpoint:
{
"mcpServers": {
"conexor": {
"url": "https://your-conexor-endpoint.conexor.io/mcp"
}
}
}
Restart Claude Desktop. That's it.
Step 4 — Ask your database anything
Open a new Claude conversation and try:
"Show me all devices that haven't synced in the last 30 days, grouped by department."
Claude will use your Conexor MCP connection to query your actual database and return real results — formatted, explained, and ready to act on.
No SQL required. No engineer interrupted.
What this unlocks
This isn't just about convenience. It's about who can now access your data.
When your IT manager can ask questions directly — without filing a ticket — you eliminate an entire category of back-and-forth. Your engineers spend time building, not translating.
And because Conexor's agent is read-only by default, you're not handing over write access to an AI. You're giving it a window, not a keyboard.
Next steps
Once you've got the basics running, try more complex questions:
- "Which applications are installed on more than 50% of our endpoints?"
- "List all devices running an OS version older than 6 months."
- "Are there any duplicate hostnames in our inventory?"
The queries get smarter as you use them. And so does your team's ability to make decisions without waiting for data.