Docs RAG
aabdelrahim/docs-rag
RAG local para documentação — busca semântica em design docs via MCP
Ver no GitHubServidor MCP que indexa documentos Markdown em ChromaDB e expõe busca semântica para Claude via MCP. Ideal para criar bases de conhecimento locais (design docs, specs, notas) consultáveis em linguagem natural sem copiar contexto manualmente.
Instalação
npx convez-ai aabdelrahim/docs-rag1 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
- Indexar e consultar documentação técnica de projetos (design docs, specs, netcode)
- Construir base de conhecimento pessoal em markdown consultável por Claude
- Recuperar respostas contextualizadas de notas sem copiar arquivos inteiros
Veredito

Implementação sólida de RAG local com MCP — embeddings via Ollama, ChromaDB persistent, integração nativa com Claude Code. Excelente para desenvolvedores que precisam consultar documentação técnica durante coding sem overhead de APIs externas.
README
A tradução para o português deste README ainda está na fila.
docs-rag
A RAG (Retrieval-Augmented Generation) MCP server that indexes a directory of Markdown files into a local ChromaDB vector store and exposes semantic search to Claude Code via the Model Context Protocol.
Built as a game design knowledge base for Esforleeg — indexes design docs, movement system specs, netcode notes, and session notes so they can be queried in plain English during development without copy-pasting files into context.
What it demonstrates
- RAG pipeline end-to-end: chunking → embedding → vector storage → retrieval
- MCP tool server exposing AI capabilities to Claude Code
- Local-first: embeddings via nomic-embed-text on Ollama, no external API calls for indexing
- ChromaDB for persistent local vector storage with cosine similarity
Architecture
Markdown files (~/infra-config/docs/)
│
▼ chunk (800 chars, 100 overlap)
nomic-embed-text (Ollama on local GPU)
│
▼
ChromaDB (~/.local/share/docs-rag/chroma/)
│
▼ cosine similarity search
MCP tools (search_docs, list_docs, reindex_docs)
│
▼
Claude Code (retrieves grounded answers from design docs)
Tools
| Tool | Description |
|---|---|
search_docs | Semantic search — returns ranked chunks with relevance scores |
list_docs | List all indexed documents |
reindex_docs | Re-index all files (run after edits) |
Example queries (in Claude Code)
- "What did I decide about stamina regeneration rate?"
- "What are the movement states and transitions?"
- "What's the rollback netcode plan for Esforleeg?"
- "Which AI portfolio project should I build next?"
Setup
python3 -m venv .venv
.venv/bin/pip install mcp chromadb httpx
# Requires Ollama running with nomic-embed-text pulled:
# ollama pull nomic-embed-text
Add to ~/.mcp.json:
{
"mcpServers": {
"docs-rag": {
"command": "/path/to/.venv/bin/python",
"args": ["/path/to/server.py"],
"env": {
"DOCS_PATH": "/path/to/your/docs",
"OLLAMA_URL": "http://localhost:11434",
"EMBED_MODEL": "nomic-embed-text"
}
}
}
}
Results
Indexes ~15 design documents (~3,500 lines total) into ~180 chunks in under 30 seconds on first run. Subsequent searches return in <100ms locally.