As AI coding agents like **Claude Code** evolve, the *real* bottleneck is no longer the model — it's **context engineering**: - How do you feed the right files? - How do you shape retrieval quality? - How do you reduce noise? - How do you keep agent inference cheap and relevant? Today, doing this well often means: ❌ running a heavy vector database ❌ gluing multiple cloud services ❌ maintaining complex pipelines ❌ spending hours configuring chunking heuristics I built **devrag** to solve exactly that. ### ✅ devrag helps Claude Code by: - Structuring codebase context intelligently - Applying opinionated chunking for agent workflows - Reducing irrelevant retrieval noise - Making embeddings ergonomic - Running everything local-first (no vendor lock) ### Why Claude Code specifically? Claude’s strength scales dramatically with **high-quality**, **precision-shaped** context. Give it shallow context, get shallow answers. Give it structured context, get *agentic reasoning*. ### With devrag you can: - Index repositories in seconds - Search by semantics + metadata - Generate high-signal context windows - Feed Claude Code what it *should* see, not everything This bridges the gap between: > “AI that can write code” and > “AI that understands your codebase.” If you’re experimenting with agents, DX tooling, or context shaping, I’d love your feedback. 🧠 Happy to answer questions all day. Thanks for being here! 🚀

DevRag

markdown vector search mcp server for claude code

markdown vector search mcp server for claude code