Before you begin

Depending on your environment, you will need:

• Java 17 or higher

  • If you're installing the AI Changelog Generator by downloading and running the .jar file directly, Java 17 or higher is required. It must be in your system PATH to be recognized in the CLI.

• Optional: VS Code 1.102 or later

• Optional: Claude Code

• Optional: Docker Desktop

Supported environments

The AI Changelog Generator can be configured in:

• VS Code — via the Command Palette using the JAR file

• Docker — via the official Docker image from Liquibase's registry

• Claude Code — via the Claude Code configuration file

What is the Liquibase AI Changelog Generator?

The Liquibase AI Changelog Generator is a local server that implements the Model Context Protocol (MCP) which is an open standard that lets AI assistants interact with external tools. When integrated with an AI agent such as Claude or GitHub Copilot, it serves as a bridge between natural language and Liquibase's changelog format.

Instead of writing:

You type:

"Create a users table with an auto-incrementing ID, a unique username up to 50 characters, and a non-null email."

The server handles the rest.

How it works

The AI Changelog Generator runs locally on your machine and integrates with your AI agent inside VS Code or Claude Desktop. When you describe a database change in natural language, your AI agent calls the appropriate MCP tool, which converts the request into a valid Liquibase changeset and returns it directly in your editor or chat interface.

Every generated changeset is automatically validated against an H2 in-memory database before it's returned to you which catches syntax errors and schema issues before they ever reach your pipeline.

The server requires no external Liquibase CLI installation. It is powered by Liquibase's native Java APIs and ships as a self-contained executable with over 300 tests.

What you can do with it

The AI Changelog Generator includes 19 specialized tools covering the full lifecycle of database schema management:

Each tool maps directly to a Liquibase Change Type and accepts natural language input. For example:

• "Add a created_at timestamp column to the orders table" → add_column

• "Create a unique index on the email column of the users table" → create_index

• "Rename the user_name column to full_name in the users table" → rename_column

Output formats

The AI Changelog Generator can generate changelogs in two formats depending on your team's workflow: XML or Formatted SQL.

XML (default)

The default output format. XML changesets are database-agnostic, work across all Liquibase-supported platforms, and let Liquibase manage rollback logic automatically at runtime. This is the best choice for teams using Liquibase across multiple databases or those who want the full range of Liquibase change type attributes.

Formatted SQL

An alternative format that produces plain .sql files using Liquibase's formatted SQL changelog syntax. SQL is generated for your specific target database, making it immediately readable by anyone familiar with SQL. Rollback statements are written explicitly into the file by the AI Changelog Generator rather than inferred at runtime. This format works well for teams with SQL-first workflows or those introducing Liquibase to an existing SQL-based process.

You can switch between formats at any time using the LIQUIBASE_MCP_OUTPUT_FORMAT environment variable in your AI Changelog Generator configuration.

Rollback handling

One of the more nuanced aspects of changelog management is rollback. The AI Changelog Generator handles this differently depending on your output format.

In XML, rollback logic is built into the Liquibase runtime and auto-generated for many Change Types. You should identify change types that require manual rollback and implement a rollback strategy for these.

In Formatted SQL, rollback must be explicitly written into the file. The AI Changelog Generator does this automatically for reversible operations (such as createTable, addColumn, and renameColumn). For destructive operations where rollback cannot be safely inferred, such as dropTable or dropColumn, the server inserts a clearly marked placeholder comment reminding you to supply the rollback SQL manually. Learn more here: Implement a rollback strategy with SQL changelogs

Next steps

Ready to set up the server? Choose your environment to get started:

• Configure the AI Changelog Generator in VS Code

• Obtain the Liquibase Changelog AI Changelog Generator Docker image

• Configure the AI Changelog Generator in Claude Desktop

• Configure the AI Changelog Generator in Claude Code