GitHub Analyzer
diegocero-dot/github-analyzer
Análise técnica completa de repositórios GitHub com Q&A via RAG
Ver no GitHubMCP server que analisa repositórios públicos do GitHub e gera relatórios técnicos detalhados sobre stack, arquitetura, débito técnico e qualidade de documentação. Integra RAG para responder perguntas em linguagem natural sobre a base de código, orquestrando agentes especializados via LangGraph e indexação em Qdrant.
Instalação
npx convez-ai diegocero-dot/github-analyzer1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Avaliar qualidade técnica e arquitetura de projetos open-source
- Gerar relatórios de débito técnico e recomendações de melhoria
- Fazer due diligence técnica antes de integrar ou contribuir a um projeto
- Explorar codebase complexa via perguntas em linguagem natural
Veredito

Ferramenta sofisticada em estágio M0 (arquitetura), com stack moderno (LangGraph, Qdrant, FastAPI, Claude API) bem escolhido para o problema. Ideal para times que precisam avaliar ou navegar codebases complexas; requer familiaridade com Python, Docker e RAG para máximo proveito.
README
A tradução para o português deste README ainda está na fila.
github-analyzer
CLI and web tool that analyzes any public GitHub repository and produces a technical report on its stack, architecture, technical debt, and documentation quality. Includes natural-language Q&A over the codebase via RAG.
Status: early development. M0 — Setup + Architecture.
What it does
Given the URL of a public GitHub repository, the tool:
- Clones (or fetches via API) the repo and indexes its code into a vector database.
- Runs a pipeline of specialized agents in parallel — stack detection, architecture analysis, technical-debt review, documentation quality — orchestrated by LangGraph.
- Synthesizes the individual reports into a structured Markdown report with numeric scores per dimension.
- Exposes a RAG channel so users can ask natural-language questions about the codebase after analysis.
Stack
- Python 3.12+ with strict type hints (
mypy --strict). - LangGraph — agent pipeline orchestration.
- Qdrant — vector DB for indexing and RAG.
- MCP server (custom) — GitHub API access with rate-limit awareness.
- FastAPI — web backend.
- Anthropic API (Claude) — primary LLM provider.
- Docker + Railway (or Fly.io) — deployment.
Full architecture in ARCHITECTURE.md.
Status
Project is in M0 — Setup + Architecture (M0.1 + M0.2 + M0.3 done). See PROJECT.md (workspace-level) for the live roadmap and milestones.
Quickstart (dev)
Requires uv (https://docs.astral.sh/uv/) and Python 3.12+.
cd code
uv sync # create virtualenv + install deps
uv run github-analyzer --help # smoke test the CLI stub
uv run pytest tests/unit/ # run the unit test suite
Set up secrets by copying .env.example to .env and filling the keys for the
milestone you are working on (see comments inside .env.example).
License
MIT — see LICENSE.
Run with Docker (FastAPI + Qdrant)
Requires Docker Desktop (or Docker Engine + Docker Compose v2).
cd code
cp .env.example .env # fill secrets you have at this milestone
docker compose up --build # starts qdrant on :6333 and api on :8000
Smoke checks:
curl http://localhost:8000/health # → {"status":"ok"}
curl http://localhost:8000/ # → service identity (name, version, status)
curl http://localhost:6333/ # → Qdrant root (collections empty in dev)
Stop the stack: docker compose down.
Wipe Qdrant data + report cache volumes: docker compose down -v.
OpenAPI docs live at http://localhost:8000/docs once the api container is up.
CLI usage (M1)
cd code
uv sync --all-extras
# Optional: copy your GitHub token (public_repo scope) into .env for higher rate limits.
echo 'GITHUB_TOKEN=ghp_xxxxxxxxxxxxxxxxxxxx' >> .env
# Fetch metadata + files + commits from a public repo.
uv run github-analyzer fetch https://github.com/octocat/Hello-World
# More commits in the listing.
uv run github-analyzer fetch https://github.com/octocat/Hello-World --commits 10
# Keep the temp clone on disk for inspection.
uv run github-analyzer fetch https://github.com/octocat/Hello-World --keep-clone
Without GITHUB_TOKEN the unauthenticated rate limit is 60 req/hour. With a token it's 5000 req/hour.