grippy-code-review¶
AI-powered PR review agent with security rule engine.
Runs with any OpenAI-compatible model as an MCP server or GitHub Actions workflow. Indexes your codebase for context-aware analysis, runs a deterministic security rule engine before the LLM, scores PRs against a structured rubric, and posts inline findings --- all as a grumpy security auditor who is reluctantly thorough.
Pages¶
- Getting Started --- Setup for MCP server, OpenAI, local LLMs, and development
- Configuration --- Environment variables, transports, and model options
- Architecture --- Modules, prompt composition, data flow
- Knowledge Graph --- Cross-PR memory, blast radius, and codebase graph
- Review Modes --- The 6 review modes and how they work
- Scoring Rubric --- How Grippy scores PRs
- Security Model --- Rule engine, codebase tool protections, and CI hardening
- Self-Hosted LLM Guide --- Run your own model with Cloudflare Tunnel
- Contributing --- Development setup, testing, and conventions
Quick links¶
- GitHub repository
- PyPI package (
pip install grippy-mcp) - MIT License
- Issue tracker