Published: 2026-02-17

Connect AI Agents to Your Database with MCP

AI coding assistants can write SQL, but they can't see your schema. They don't know your table names, column types, or relationships. The Model Context Protocol (MCP) changes that — it gives AI agents a secure, structured way to interact with live databases. This guide shows you how to set it up.

1. What Is MCP and Why Do SQL Developers Need It?

The Model Context Protocol (MCP) is an open standard for connecting AI tools to external data sources. Think of it as a universal adapter between AI assistants and the applications they need to work with.

For SQL developers, MCP solves a fundamental limitation: AI tools are great at writing SQL syntax, but they have no idea what your actual database looks like. Without schema context, AI-generated queries reference tables that don't exist, use wrong column names, and miss relationships between tables.

MCP changes the workflow from "copy-paste your schema into a chat window" to "the AI already knows your schema, table structures, and can execute queries directly."

2. The Problem: AI Can't See Your Schema

You've probably experienced this workflow:

1. Ask an AI assistant to write a query
2. Get back syntactically correct SQL that references users when your table is called dbo.Customers
3. Paste your schema definition into the chat
4. Get a better query, but it misses a foreign key relationship
5. Paste more context
6. Repeat

This manual context-passing is slow, error-prone, and doesn't scale. A database with 50 tables and 200 stored procedures can't be summarized in a chat message. Even if you could, the context would be stale the moment someone adds a column.

MCP eliminates this friction. Instead of manually describing your database, the AI agent connects to a live data source and introspects the schema in real-time.

3. How MCP Bridges AI Tools and Live Databases

Here's how MCP works with Jam SQL Studio:

1. Jam SQL Studio connects to your databases — SQL Server, PostgreSQL, MySQL, SQLite, Azure SQL
2. Jam SQL Studio runs a local MCP server — a lightweight HTTP endpoint on localhost
3. Your AI tool connects to the MCP server — using a bearer token for authentication
4. The AI agent calls MCP tools — to list connections, introspect schemas, execute queries, and control the UI

The AI tool never connects directly to your database. Jam SQL Studio acts as a secure intermediary that manages connections, enforces permissions, and logs every action.

4. Step-by-Step: Connect Claude Code to Your Database

Claude Code has native HTTP support for MCP servers, making it the simplest setup.

{
  "mcpServers": {
    "jam-sql": {
      "type": "http",
      "url": "http://127.0.0.1:14507/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN_HERE"
      }
    }
  }
}

{
  "mcpServers": {
    "jam-sql": {
      "type": "http",
      "url": "${JAM_SQL_MCP_ENDPOINT:-http://127.0.0.1:14507/mcp}",
      "headers": {
        "Authorization": "Bearer ${JAM_SQL_MCP_TOKEN}"
      }
    }
  }
}

claude mcp add --transport http jam-sql http://127.0.0.1:14507/mcp \
  --header "Authorization: Bearer YOUR_TOKEN_HERE" \
  --scope user

5. Step-by-Step: Connect Claude Desktop

Claude Desktop uses stdio transport, so it requires the mcp-remote proxy and Node.js 20+.

node --version  # Must be v20.0.0 or higher

{
  "mcpServers": {
    "jam-sql": {
      "command": "/path/to/node20+/bin/npx",
      "args": [
        "-y",
        "mcp-remote",
        "http://127.0.0.1:14507/mcp",
        "--header",
        "Authorization:${JAM_SQL_TOKEN}"
      ],
      "env": {
        "JAM_SQL_TOKEN": "Bearer YOUR_TOKEN_HERE"
      }
    }
  }
}

6. Step-by-Step: Connect Codex CLI

OpenAI's Codex CLI uses TOML configuration for MCP servers.

[mcp_servers.jam-sql]
url = "http://127.0.0.1:14507/mcp"
bearer_token_env_var = "JAM_SQL_MCP_TOKEN"

export JAM_SQL_MCP_TOKEN="YOUR_TOKEN_HERE"

7. What Can AI Do Once Connected?

Once your AI agent is connected via MCP, it can interact with your databases through Jam SQL Studio's tools. Here's what becomes available:

8. Security: Why This Is Safe

Giving AI access to your database sounds risky. Here's why Jam SQL Studio's approach is designed to be safe:

9. Real-World Use Cases

Debugging Application Issues

Ask your AI agent to check the state of your data while debugging. Instead of switching between your IDE and a SQL client, the AI can query relevant tables, inspect recent records, and correlate data across multiple tables — all from your terminal.

Query Optimization

Paste a slow query into Claude Code and ask it to optimize. With MCP, it can generate the actual execution plan, identify table scans and missing indexes, suggest improvements, and run the optimized version to compare performance — all without you touching the SQL IDE.

Schema Documentation

Ask the AI to document your database schema. It can introspect every table, view, and stored procedure, then generate comprehensive documentation with descriptions, relationships, and data type summaries.

Data Analysis and Reporting

Describe the report you need in natural language. The AI can write the SQL, execute it, read the results, and iterate until the output matches your requirements. It can also configure charts directly in Jam SQL Studio to visualize the data.

Building Migration Scripts

The AI can compare your current schema against your application models and generate migration scripts. With read access to the actual database, the migrations account for existing data, constraints, and dependencies.

10. FAQ

Which databases does MCP support?

MCP works with any database supported by Jam SQL Studio: SQL Server, Azure SQL, PostgreSQL, MySQL, MariaDB, and SQLite. The AI agent connects to Jam SQL Studio, not directly to the database, so no database-specific configuration is needed on the AI side.

Do I need a paid plan?

The MCP server is available in the free Personal tier. AI Workspace features (file sync, schema export, auto-generated context files) require Pro. The AI Chat Sidebar uses your existing Claude Code installation at no additional cost.

Can I use this with tools other than Claude?

Yes. Any tool that supports MCP or HTTP/JSON-RPC can connect: Codex CLI, OpenCode, GitHub Copilot (via MCP extension), LangChain agents, AutoGen agents, or custom scripts using curl or any HTTP client.

What if I don't want the AI to run queries?

Set the permission level to Block (the default). The AI can still introspect schemas, open query tabs, and write SQL into the editor — but it cannot execute any queries. You run them manually when ready.

Can other people on my network access the MCP server?

No. The server binds exclusively to 127.0.0.1 (localhost). It does not listen on any network interface. Only processes on your local machine can reach it.