# Caelune [Health: Active]

**Category:** 🔎 end to end RAG platforms  
**Repository:** https://github.com/EllisMorrow/Caelune  
**GitHub Stars:** 45  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/caelune

## Description
Read-only local-first MCP server for private Markdown, PDF, and Tika-backed search on Windows.

## Claude Desktop Quick Installation
Heuristic fallback — verify the package name and runner against the repository README before running it. Uses `npx` (confidence: low):

```json
"mcpServers": {
  "caelune": {
    "command": "npx",
    "args": ["-y","caelune"]
  }
}
```

## Documentation

## What the Caelune MCP server does

The Caelune MCP server gives MCP-capable AI clients read-only access to knowledge bases stored on a Windows computer. It searches content indexed by the Caelune desktop application and returns results with source labels so the originating document type and location can be reviewed.

Supported content includes Markdown, PDF, and additional formats handled through Apache Tika extensions. Markdown processing includes awareness of common Logseq properties, block references, and embeds. The server is intended for retrieval, not for managing the underlying collection: it cannot build knowledge bases, remove indexed data, or change application configuration.

## How it works

The desktop app must prepare at least one knowledge base before the MCP connection can provide useful search results. Caelune reads source files locally and creates separate indexes for its supported document sources. The retrieval pipeline combines SQLite FTS5 lexical matching, structure-aware scoring, LanceDB vector search, and optional reranking. Its local models include `BAAI/bge-m3` for semantic retrieval and optionally `BAAI/bge-reranker-v2-m3` for reranking.

After indexing, the MCP executable exposes the retrieval core through a local `stdio` connection. An MCP client can first inspect server readiness and active retrieval mode, then submit a search request. The desktop app can also use live watch to update indexes after source files change, but the MCP package itself does not build or modify those indexes.

## Setup and configuration

Use the Windows desktop release to choose a local data root, add Markdown knowledge-base directories, install or repair the required CPU or CUDA Runtime and local models, and perform an initial full build. The release also provides an MCP-specific Windows ZIP for manual `stdio` configuration and an `.mcpb` package for Registry or MCPB-aware clients.

Configure the MCP client to launch `Caelune-MCP.exe` from the downloaded package. Some clients do not accept local `stdio` servers directly and may need a local proxy. The application requires 64-bit Windows and sufficient storage for its application files, external Runtime, models, and indexes.

## Tools and capabilities

The Caelune MCP server exposes two read-only tools:

- `omniclip.status` checks readiness and reports the active retrieval mode.
- `omniclip.search` retrieves source-labelled search results.

Search can use the indexes built from Markdown, PDF, and Tika-backed sources. Caelune’s desktop interface provides additional controls for query scope, retrieval settings, result review, page filtering, and context selection, while the MCP interface is limited to the documented read-only operations.

## Limitations and notes

Caelune is local-first, but it may access the network when the user explicitly downloads Runtime components, models, Java, or Tika. Retrieved text leaves the application when it is copied, exported, or requested by an MCP client, and any connected AI service applies its own privacy policies.

The MCP server does not create an index on its own, so a desktop-app build is required first. CPU execution is supported; a compatible NVIDIA GPU is optional. Actual Tika compatibility depends on the source file and parser even though the Tika catalog lists many format entries. Keep source backups and review retrieved context before relying on it for high-stakes work.

_Full upstream README: https://allmcps.com/mcp/caelune/readme_