GitHubnpm

vibe-hnindex

vibe-hnindex is a local Model Context Protocol (MCP) server that lets AI assistants search your codebase. Index a project once, then search it in every AI session — your code stays on your machine.

What It Does

vibe-hnindex builds a searchable index of your codebase using SQLite (for keyword and FTS5 full-text search) and Qdrant (for semantic vector search powered by Ollama embeddings).

Once indexed, AI tools can:

  • Search your code by keyword, natural language, or hybrid (both combined)
  • Find definitions of functions, classes, and symbols
  • Use regex to find patterns across your entire codebase
  • Get smart context for tasks, questions, or refactoring
  • Benchmark search performance across different modes

Key Features

🔍 Multi-Mode Search

Keyword (FTS5+BM25), semantic (Qdrant vectors), hybrid (RRF fusion), regex, and symbol search.

⚡ Hyper-Speed Indexing

Single-pass indexing with parallel workers (~30-40% faster since v0.9.1).

📦 Incremental Updates

SHA-1 hashing — only re-index changed files.

🔒 100% Local

Code never leaves your machine. SQLite + Qdrant run locally.

🧠 Smart Context

Auto-detect task types and gather relevant context for AI agents.

🌐 Streaming Search

Parallel keyword + semantic search with progress updates.

How It Works

  1. Index: You point vibe-hnindex at a directory. It scans files, chunks them (~60 lines), embeds chunks via Ollama, and stores everything in SQLite + Qdrant.
  2. Search: AI assistants call the search tool, which queries the index using keyword, semantic, or hybrid modes.
  3. Results: Ranked code snippets with file paths and line ranges are returned to the AI for context.

Supported Platforms

vibe-hnindex works with any MCP-compatible AI tool:

Claude DesktopClaude CodeCursorWindsurfVS Code CopilotGoogle AntigravityContinue.dev

Requirements

  • Node.js ≥ 20 (LTS recommended)
  • Ollama — for embeddings (required for semantic/hybrid search)
  • Qdrant — Docker or Qdrant Cloud (optional; keyword search works without it)