# oomkapwn/enquire-mcp [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/oomkapwn/enquire-mcp  
**GitHub Stars:** 34  
**npm Downloads (last month):** 1867  
**Views:** 10  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/oomkapwn-enquire-mcp

## Description
Long-term memory for AI agents (Claude Code/Desktop, Cursor, ChatGPT, Codex, OpenClaw) backed by a local Obsidian markdown vault. Hybrid retrieval (BM25 + ML embeddings + BGE reranker, RRF-fused), HNSW + int8 quantization, agentic RAG (HyDE + sub-question), GraphRAG-light (Louvain wikilink community detection), standalone Obsidian Bases, PDFs + Tesseract OCR. 46 tools, 19 MCP prompts, MIT, SLSA L2, zero cloud calls during serve. npx -y @oomkapwn/enquire-mcp serve --vault

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "enquire-mcp": {
    "command": "npx",
    "args": ["-y","@oomkapwn/enquire-mcp"]
  }
}
```

## Documentation

## What oomkapwn/enquire-mcp does

oomkapwn/enquire-mcp exposes an Obsidian vault through the Model Context Protocol so agents can search and inspect knowledge stored in files they already maintain. Markdown notes, frontmatter, wikilinks, PDFs, and OCR-derived text can participate in retrieval. Results preserve source paths, and PDF matches retain page references for citation. Search responses can also include document age and a stale indicator, helping an agent decide when to verify older material.

The server is designed for shared, vendor-neutral agent memory. The source files remain ordinary, editable Obsidian files rather than being moved into a proprietary conversation-memory store. Claude Code, Claude Desktop, Cursor, ChatGPT, Codex, OpenClaw, and other MCP-compatible agents can use the same vault through their MCP connections.

## How it works

At serving time, the vault is indexed locally. Retrieval combines lexical matching with machine-learning embeddings, then can apply a BGE cross-encoder reranker. Reciprocal rank fusion combines signals, while HNSW indexing and int8 quantization support the dense retrieval path. Optional recency weighting can favor newer notes, and agent-oriented retrieval includes HyDE and bounded sub-question generation.

The system also adds a lightweight graph dimension by detecting wikilink communities with Louvain clustering. Search output can expose per-signal scores and stage information, which gives agents more context about why a result was selected. PDFs follow the same local retrieval path as Markdown, with Tesseract OCR available for scanned pages.

## Setup and configuration

Install the published npm package globally:

```bash
npm i -g @oomkapwn/enquire-mcp
```

Start the server by supplying the vault path:

```bash
npx -y @oomkapwn/enquire-mcp serve --vault /path/to/Obsidian-Vault
```

Claude Code can register the same command with its MCP configuration:

```bash
claude mcp add obsidian -- npx -y @oomkapwn/enquire-mcp serve --vault /path/to/Obsidian-Vault
```

The vault location is passed through the `--vault` command-line option. Published GitHub releases and published npm versions are the appropriate sources for usable artifacts.

## Tools and capabilities

- Search Markdown and PDF content with hybrid lexical and semantic retrieval.
- Preserve note paths, PDF page citations, age metadata, and stale status in search results.
- Parse Obsidian Canvas files as typed graph data.
- Run the supported Dataview-style `LIST` and `TABLE` query subset.
- Execute supported Obsidian Base filters.
- Use OCR to retrieve text from scanned PDF pages.
- Apply optional BGE reranking, recency weighting, HyDE, and sub-question retrieval.
- Inspect retrieval signals and stage-level scoring details.

## Limitations and notes

The server is read-only by default. Write tools become available only when it is started with `--enable-write`, so deployments that need mutation should enable that capability deliberately. Dataview and Obsidian Base support covers documented subsets rather than every feature of those systems.

Processing is local during serving, and enquire-mcp initiates no outbound calls or telemetry. Any later handling of returned context depends on the MCP client and model connected to it. The project is distributed under the MIT license.

_Full upstream README: https://allmcps.com/mcp/oomkapwn-enquire-mcp/readme_