Skip to content
TwinCommitDocs

TwinCommit documentation

TwinCommit is a log-based change data capture (CDC) engine. It reads the transaction log of your source database and applies every committed transaction exactly once, in commit order, to the target you choose. It runs as a single binary and installs nothing in the source database.

These docs cover installation, configuration, every supported source and target, the CLI, day-two operations and the TwinCommit MCP server.

Start here

Section titled “Start here”
QuickstartReplicate a Db2 table into PostgreSQL in about five minutes.
InstallationLinux binary, Docker, Helm chart or Homebrew.
ArchitectureHow the log reader, transaction buffer and apply loop fit together.
Configuration referenceEvery key in a pipeline YAML file.

Sources and targets

Section titled “Sources and targets”

Operate with your AI assistant

Section titled “Operate with your AI assistant”
MCP server overviewManage pipelines from Claude, GitHub Copilot or Cursor.
Install the MCP serverOne command for Claude Code; config snippets for every other client.

Core guarantees

Section titled “Core guarantees”
  • Nothing installed on the source. No triggers, change tables or stored procedures. See sources.
  • Exactly-once delivery. The checkpoint commits in the same target transaction as the data. See exactly-once.
  • Commit order. Transactions are applied whole, in the order they committed. Rolled-back work never reaches the target.
  • Fault isolation per table. An operation the log cannot describe affects one table, not the pipeline. See broken tables.