Introduction
RubberDuck is a software truth layer for AI-assisted engineering. While every AI coding tool generates code, RubberDuck lets your agent understand the code it’s writing.
It builds a semantic model of the indexed repository, definitions, dependencies, call chains, and data flows, and feeds that knowledge to your agent on demand. The model lives outside the context window, so your agent stays grounded even as the project grows.
Supports
Via MCP
Start here
Quickstart with MCP
Install RubberDuck and connect your first repository.
Connect a repository
Add your GitHub repository so every pull request is reviewed automatically.
CLI Reference
The full @rubberduck command surface, grouped by workflow.
What you can do with RubberDuck
Understand a codebase
Trace call chains and data flow to find the right place to start.
Investigate security
Audit sensitive paths and get a verified verdict, not a guess.
Trace a bug
Find the root cause and confirm the fix actually holds.
Review a PR / change
Catch what a diff-only reviewer would miss.
Generate a patch
Propose a committable fix, gated on verify.
Plan a feature
Scope the change and its blast radius before you write a line of code.
Why trust RubberDuck vs. existing AI code review tools?
A valid result always names:
- The repository, branch or ref
- A pinned commit SHA
- Exact file paths
- Supporting evidence
If that evidence can’t be produced:
RubberDuck reports the result as blocked or incomplete, never an unsupported clean, fixed, or merge-ready claim.
You won’t get a hallucinated result.