GETTING STARTED / v0.16.1
Installation
Start with local workspace context and multilingual code navigation. Add semantic retrieval when your workflow needs it.
Requirements
- Node.js 22 or newer and npm.
- An MCP client such as Codex, Claude Code, Cursor, VS Code, Windsurf or Antigravity.
- A readable project directory. SQLite storage defaults to
~/.vibe-hnindex.
Workspace context and index_code_graph do not require Qdrant, Ollama or a cloud API key. AST graph analysis uses the JS/TS compiler plus bundled Tree-sitter profiles for additional languages.
Connect your editor
cd /your/project
npx -y hnindex-cli init --mcp codex
npx -y hnindex-cli init --mcp claude
# Other project-scoped targets
npx -y hnindex-cli init --mcp cursor-project
npx -y hnindex-cli init --mcp vscodeProject-scoped targets bind HNINDEX_PROJECT_ROOT automatically. For a global configuration, either use current MCP client roots or bind a root explicitly:
npx -y hnindex-cli init --mcp windsurf --project-root /your/projectRestart or reconnect your MCP client after updating its configuration. Follow the quick start to initialize the local graph.
Codex
The Codex target writes project .codex/config.toml and binds the workspace automatically. Codex loads this file only for trusted projects. Restart or reconnect Codex after setup.
npx -y hnindex-cli init --mcp codex
npx -y hnindex-cli init-skill --target codexThe optional skill is installed in .agents/skills/use-vibe-hnindex. See the Codex MCP setup for TOML and connection checks.
Optional: semantic and hybrid retrieval
Full codebase vector indexing and semantic/hybrid search use Qdrant plus your chosen provider: Ollama, OpenAI, Voyage, Gemini or an OpenAI-compatible endpoint.
docker run -d --name hnindex-qdrant -p 6333:6333 \
-v hnindex-qdrant:/qdrant/storage qdrant/qdrant
# For local embeddings only (Ollama installed separately):
ollama pull bge-m3:567mConfigure EMBEDDING_PROVIDER and the matching key/model in the MCP server environment. See configuration for settings. Cloud embeddings send code chunks to that provider; Ollama can keep embedding traffic local.
Verify the connection
Ask the agent to call workspace_context with the project path, then index_code_graph. Use diagnostics when troubleshooting optional embedding and Qdrant services. Do not treat a missing optional service as a prerequisite for offline graph navigation.