Oracle announced it evaluated MCP and has integrated it into its core developer tools, making the Oracle Database immediately available on any platform supporting MCP.
This integration is made available via our modern command line to the Oracle Database, Oracle SQLcl.
Oracle SQLcl is the command-line interface (CLI) for Oracle Database and is shipped with popular tools such as the Oracle SQL Developer extension for VS Code.
It can now be run as an MCP Server and provides MCP tools that allow an AI assistant to securely connect to Oracle Database. SQLcl manages the credentials on the end user’s machine and runs SQL and PL/SQL queries and scripts.
Oracle recommends taking extreme caution when granting the LLM direct access to production databases; instead, use a sanitized, read-only replica or a dedicated data subset. Regularly audit queries executed by the LLM to detect anomalies or attempts to access restricted data.
Oracle Database professionals and administrators should ensure that the database connections utilized by the MCP Server are defined with database users having the least required privileges to accomplish the work at hand. This will limit what is accessible to the LLM.
Now that the Oracle SQLcl MCP server has been made available to LLMs, developers can use AI agentic workflows to directly run generated SQL statements on an Oracle Database and interact with the results.
Contrast this with simply having your AI provide SQL for you to manage. Now, Oracle offers developers a workflow where the AI agent can implement its advice directly on the Oracle Database and evaluate the responses.
Model Context Protocol (MCP) was introduced in November 2024 by Anthropic. Until MCP, large language models (LLMs) have never been able to directly interact with external data sources and APIs. Instead, the application layer was responsible for orchestrating context augmentation by using the model to enrich user questions with additional information.
MCP's explosive growth comes from the standardization of a protocol used to add context to an LLM. The absence of such a protocol meant that every tool, every API, and so on, would have to create a bespoke implementation for each platform or integration to offer the same functionality.
For more information about this news, visit www.oracle.com.