Selvage

Brings AI-powered code review into your MCP-enabled editor by analyzing Git diffs with AST-based context extraction. Exposes `perform_review` for direct LLM analysis using OpenRouter models (Claude, GPT, Gemini) and `get_review_context` for agent-delegated reviews that let your host agent handle the actual review using structured diff and context data. The latter means no API key required if you're just extracting review context for Claude Code or Cursor to process. Supports reviewing staged changes, unstaged work, branch comparisons, or specific commits. Uses tree-sitter to pull minimal relevant code blocks and their dependencies, keeping context lean. Automatically splits large reviews across multiple turns when you hit token limits. Reach for this when you want instant code review feedback without waiting for human reviewers or context-switching to a separate tool.