AI Code Review Server
A developer-tooling project that reviews code changes with LLMs, structured risk scoring, Prometheus metrics and an MCP-compatible integration path.
Outcome: Turns LLM review output into validated severities, categories and risk scores; includes dashboard, metrics and no stored credentials.
TypeScriptExpressReactViteZodDockerMCPPrometheus
Completed January 2026
AI Code Review Server
A code review assistant that analyzes source code or diffs, returns structured feedback, highlights risk areas and exposes an MCP-compatible integration path for AI-enabled developer workflows.
What it does
- Accepts code snippets or diffs through a React dashboard.
- Sends review requests to LLMs through OpenRouter.
- Returns issues grouped by severity and category.
- Produces a risk score to help prioritize follow-up.
- Supports a protocol-oriented integration path for AI coding tools.
- Exposes Prometheus metrics and structured logging for operational visibility.
- Avoids stored credentials by accepting the API key per request.
Technical decisions
- TypeScript end-to-end for shared mental models across frontend and backend.
- Zod validation to keep model responses and API payloads safe at runtime.
- Express API for a small, predictable service surface.
- Docker Compose for quick local setup and reproducible environments.
Why it matters
This project explores how AI can become useful only when wrapped in structure: validation, deterministic response formats, clear severity levels and a workflow that developers can trust.