v0.1.0 — Phase 1: MVP

Phase 1 delivers a complete local PDF RAG system exposed as an MCP server, installable from PyPI and ready to use in Cursor.

Core RAG pipeline

• PDF loading — Load PDFs with pypdf, handle layout quirks, extract per-page text.
• Chunking — Configurable chunk size/overlap (via env), metadata preserved.
• Vector store — Persistent Chroma store on disk; embeddings stored and queried.
• RAG chain — Retrieve chunks, generate answers with citations; test-covered.

Indexing & querying

• index_pdf — Replaces existing chunks for the same document (no duplicates).
• Query — End-to-end query support over the index, including smoke tests.

MCP server

• Tools — FastMCP server: query_pdf, add_pdf, list_pdfs, remove_pdf.
• CLI — oracle-rag-mcp entry point; env loaded from config/cwd.
• Errors — Tool failures logged to stderr; structured errors propagated to clients.
• Logging — Noisy INFO/WARNING from MCP, Chroma, OpenAI, httpx, pypdf suppressed for readable Cursor output.

Configuration & persistence

• Env — .env from ~/.config/oracle-rag/, ~/.oracle-rag/, then cwd; default persist dir ~/.oracle-rag/chroma_db.
• Config — Chunk size/overlap with validation and tests.

Distribution

• PyPI — Published as oracle-rag with oracle-rag-mcp entry point.
• Cursor — MCP config: command: oracle-rag-mcp; uses ~/.oracle-rag for .env and chroma_db.
• CI — GitHub Actions: build and publish to PyPI on v*.*.* tags.

Testing

• Coverage — PDF loading, chunking, embeddings, vector store, RAG chain, config, MCP tools, error handling, duplicate PDF replacement.