Using Oracle SQLcl MCP Server with Oracle 19c: A Step-by-Step Guide for NLP-Based Database Queries

 

Introduction

With the rapid evolution of AI, databases are no longer limited to traditional SQL-only interactions. Oracle has taken a major step forward by introducing MCP (Model Context Protocol) support in SQLcl, allowing AI tools like Claude Desktop to interact directly with Oracle databases using natural language.

In this blog, I’ll walk you through a hands-on, end-to-end setup of Oracle SQLcl MCP Server with an on-prem / OCI-hosted Oracle 19c database, and show how conversational AI can query enterprise databases securely.

This guide is ideal for Oracle DBAs, Cloud Architects, and AI-curious professionals who want to explore NLP-driven database access.

   Image source:-https://blogs.oracle.com/database/introducing-mcp-server-for-oracle-database

---

Architecture Overview

AI Client (Claude Desktop)
⬇️ MCP Protocol
SQLcl MCP Server (Local Machine)
⬇️ JDBC
Oracle Database 19c (OCI / On-Prem)

The AI never connects to the database directly. SQLcl acts as a secure MCP bridge, translating natural language into database operations.

---

Prerequisites

Before starting, ensure you have:

• Oracle Database 19c (On-Prem or OCI Compute VM)

• Windows laptop or desktop

• Internet access to download tools

• Basic Oracle SQL knowledge

---

Step 1: Install JDK 17 (Required for SQLcl)

Oracle SQLcl requires Java 17.

• Download JDK 17 for Windows from Oracle

• Install using the .exe

• Set JAVA_HOME and update PATH

Verify:

java -version

---

Step 2: Install Oracle SQLcl

• Download SQLcl from Oracle

• Unzip it to a directory (example):

  C:\AI\sqlcli

SQLcl is portable—no installer required.

---

Step 3: Install Claude Desktop

Claude Desktop will act as the AI MCP client.

• Download Claude Desktop

• Install and launch once

• Close it before MCP configuration

---

Step 4: Prepare Oracle Database 19c

Verify PDBs

show pdbs;

Ensure your PDB (e.g., ORCLPDB) is in READ WRITE mode.

Listener and Network Setup

• Ensure port 1521 is open

• Disable firewall (lab use only):

systemctl stop firewalld

systemctl disable firewalld

• Confirm connectivity from Windows:

Test-NetConnection <DB_PUBLIC_IP> -Port 1521

---

Step 5: Create SQLcl Connection

Launch SQLcl:

sql /nolog

Create and save a connection:

conn -save oracle19c_mcptest -savepwd system/password@<IP>:1521/ORCLPDB

Validate:

CONNMGR test oracle19c_mcptest

---

Step 6: Start SQLcl MCP Server

sql -mcp -name oracle19c_mcptest

You should see:

MCP Server started successfully

This process must remain running.

---

Step 7: Configure Claude Desktop for MCP

Edit Claude configuration file:

{

"mcpServers": {

"oracle19c": {

"command": "C:/AI/sqlcli/sqlcl/sqlcl/bin/sql.exe",

"args": ["-mcp", "-name", "oracle19c_mcptest"]

}

}

}

Restart Claude Desktop and allow MCP access when prompted.

---

Step 8: Follow Least Privilege (Best Practice)

Instead of SYSTEM, create an application user:

CREATE USER app_user IDENTIFIED BY password;

GRANT CREATE SESSION, CREATE TABLE TO app_user;

Create sample data:

CREATE TABLE sales_orders (...);

INSERT INTO sales_orders VALUES (...);

COMMIT;

Create a separate SQLcl MCP connection for this user.

This ensures:

• AI only sees approved schemas

• SYS/SYSTEM access is avoided

---

Step 9: Test NLP Queries via Claude

Now the magic ✨

Ask Claude:

Claude:

• Understands intent

• Calls SQLcl MCP

• Executes SQL

• Returns results

No SQL typing required.

---

Security Considerations

✔ SQLcl connections are local-only ✔ Credentials stored in user profile ✔ Secure with OS file permissions ✔ Use separate DB users ✔ Optional: Oracle Wallet for credentials

AI never gets raw database access.

---

Why This Matters

This setup demonstrates:

• Conversational AI for ad-hoc querying

• AI + Oracle DB without exposing credentials

• Perfect for DBAs, Support, and Architects

---

Final Thoughts

Oracle SQLcl MCP Server bridges the gap between enterprise databases and modern AI—securely, locally, and powerfully.

If you’re running Oracle 19c today, you can already start experimenting with conversational data access.