Marx spins up multiple AI agents in Docker containers, each reviewing your PR independently. It merges their findings, removes duplicates, and gives you a pending GitHub review you can edit before submitting.
- Parallel multi-agent reviews with automatic deduplication
- YOLO-mode agents run in Docker with review tools and PR context
- Create a pending GitHub review from merged issues (edit before submit)
- Structured review outputs with a merged summary
- Interactive PR selection (excludes your own PRs and PRs assigned to you)
- Works with local CLI configs or API keys (config dirs are mounted into containers)
- JSON output for automation via
--json-output
git- gh (authenticated)
docker
uv tool install marx-aiNeed Nix or a source install? See Installation.
Create ~/.marx with your GitHub token and any agent keys you want to use.
The GitHub token must be a classic PAT with the repo scope
(details):
cat >~/.marx <<'MARX'GITHUB_TOKEN=ghp_your_token_hereANTHROPIC_API_KEY=your_claude_keyOPENAI_API_KEY=your_openai_keyGEMINI_API_KEY=your_gemini_keyMARXIf you already use the agent CLIs locally, Marx copies ~/.claude, ~/.codex, and ~/.gemini
into the containers so those configs work there too. API keys are optional.
See Configuration for details.
# Interactive PR selection
marx
# Review a specific PR with all agents
marx --pr 123 --agents claude,codex,gemini
# Machine-readable output
marx --pr 123 --json-output