Review code like magic.
MagicReview is a local-first, self-hostable AI code review platform for CLI, Docker, Dashboard, MCP tools, and GitHub App workflows.
MagicReview defaults are intentionally conservative and offline by default:
- Runs local static analysis by default.
- Does not call real LLM providers by default.
- Does not upload source code unless network access is explicitly enabled.
- Uses
MGREVIEW_environment variables and.magicreview/data by default.
pip install magicreview mgreview --help mgreview project examples/multi_agent_project --agents --format terminal
From source:
git clone https://github.com/KrisMagical/review-agent.git cd review-agent pip install -e ".[all,dev]" mgreview --help
mgreview --help
mgreview --version
mgreview file app/main.py
mgreview diff
mgreview project .
mgreview project examples/multi_agent_project --agents --format json
mgreview dashboard init-db
mgreview dashboard serve
mgreview mcp
mgreview github-app serveOptional standalone entry points are also packaged for service runners:
mgreview-mcp mgreview-dashboard mgreview-github-app
docker build -t magicreview . docker run --rm magicreview mgreview --help docker run --rm -v "$PWD:/workspace" magicreview mgreview project /workspace --format json docker compose up dashboard docker compose up github-app
Docker Compose persists SQLite data in ./.magicreview.
MagicReview uses MGREVIEW_ as its environment variable prefix:
MGREVIEW_DB_PATH=.magicreview/magicreview.db MGREVIEW_LLM_PROVIDER=none MGREVIEW_NETWORK_ENABLED=false MGREVIEW_ALLOW_LLM=false MGREVIEW_CODE_SHARING_MODE=none MGREVIEW_AUTH_ENABLED=false
Default data lookup order:
MGREVIEW_DB_PATH.magicreview/magicreview.db
MagicReview supports diff_only and full_project review modes. The generated pull request summary ends with:
Generated by MagicReview.
MagicReview is released under the MIT License. See LICENSE.