AI code review for the CLI, CI, and MCP hosts. Review staged changes locally, gate PRs in CI, or drive the engine from any MCP-capable agent (Claude Code, Codex, and others). Deterministic engine + LLM, stable JSON envelope on stdout.
Docs: cr.miu.sh
# macOS / Linux — detects OS/arch, verifies checksum:
curl -fsSL https://cr.miu.sh/install.sh | sh
# Homebrew:
brew install vanducng/tap/miucr
# From source (Go 1.25+):
go install github.com/vanducng/miu-cr/cmd/miucr@latestWindows: download miucr_windows_x86_64.zip from Releases.
No API key? Use your ChatGPT plan:
miucr login --provider openai # browser login, caches a token
miucr review --stagedBring your own key:
export ANTHROPIC_API_KEY=... # or OPENAI_API_KEY
miucr review --staged # review staged changes
miucr review --from main --to HEAD --gate high # range; exit 2 on high+ finding
miucr review --pr owner/repo#123 --post # GitHub PR with inline commentsFull guide: Getting started
- Local review — staged changes, commit range, or a single commit
- GitHub PR review — inline comments, one upserted summary, head-SHA anchoring, fork-safe
- CI / GitHub Action — drop-in reusable action; see
examples/workflows/ - Project rules —
.miu/cr/rules/*.mdglob-scoped context injected per file - Evaluation —
miucr evalcompares reviewer commands against expected findings - MCP server —
miucr mcpexposesreview_run/review_getover stdio
Full reference: cr.miu.sh
go build ./cmd/miucr # single static binary (CGO_ENABLED=0)
go test ./... # table tests + fakes, no live network or LLMAgent skill: .agents/skills/miucr/SKILL.md — agent-agnostic; .claude/skills/miucr is a symlink to it.
Apache-2.0. See LICENSE.
