# lumen

Indexes your project with Go AST parsing and finds code by meaning instead of reading whole files. Runs locally, no cloud and no API keys.

- plugin · `type=plugin` · Claude Code · ★ 312
- Author: Ory Corp
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/ory/lumen
- Updated: 2026-10-11
- HTML: https://affpapa.org/en/nevibecode/plugins/lumen

## What it does

Lumen is an MCP server for semantic code search. It parses files with Go AST, computes embeddings on a local model, and exposes a `semantic_search` tool to the agent. Claude gets back only the fragments that match the query, not entire files.

You need Ollama running with the embedding model pulled first: `ollama pull ordis/jina-embeddings-v2-base-code`. No local embedding server means no search, so check that before anything else. The plugin then downloads its binary from GitHub releases, indexes the project in the background, and registers the MCP tool. Indexing uses Merkle tree change detection, so later runs only touch files that changed.

Alongside `semantic_search` there are `health_check` and `index_status` tools, plus two shared skills: `doctor` for a health check and `reindex` for forced re-indexing. In Claude they run as `/lumen:doctor` and `/lumen:reindex`. Linux, macOS and Windows are supported.

## Install

### Claude Code (native platform)

**Plugin marketplace**
In the agent chat (not a shell):
```text
/plugin install lumen@claude-plugins-official
```
In the agent chat (not a shell):
```text
/plugin marketplace add anthropics/claude-plugins-community
```
In the agent chat (not a shell):
```text
/plugin install ory-lumen@claude-community
```
In the agent chat (not a shell):
```text
/plugin marketplace add anthropics/claude-plugins-official
```

## When to use it
- find a function from a description without opening dozens of files
- cut token spend on reading through a codebase
- re-index a project after a large refactor
- verify that the MCP server and embedding model actually work
- search code without shipping sources to a cloud service

## Example prompts
- use semantic_search to find where request authorization is handled
- run /lumen:doctor and tell me if indexing is healthy
- force a re-index with /lumen:reindex

## Similar
- [rag-engineer](https://affpapa.org/en/nevibecode/skills/sickn33-rag-engineer.md) — skill · 1000 installs
- [recall](https://affpapa.org/en/nevibecode/skills/parcadei-recall.md) — skill · 493 installs
- [embedding-strategies](https://affpapa.org/en/nevibecode/skills/embedding-strategies.md) — skill · 11391 installs
- [similarity-search-patterns](https://affpapa.org/en/nevibecode/skills/similarity-search-patterns.md) — skill · 10700 installs
- [rag-architect](https://affpapa.org/en/nevibecode/skills/httpsgithubcomjeffallan-rag-architect.md) — skill · 4700 installs
- [upstash-vector-js](https://affpapa.org/en/nevibecode/skills/upstash-vector-js.md) — skill · 1300 installs

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md