MCP 服务器

EzRAG MCP Server

Provides semantic search and keyword search over Obsidian notes, along with direct note retrieval, allowing external AI agents to query and access the vault.

README

EzRAG – AI-Powered Search for Obsidian Notes

EzRAG turns your Obsidian vault into a Gemini File Search index so you can run semantic search, chat over your notes, and expose your vault through MCP tools. Everything stays in your Google account; the plugin simply keeps the index up to date.

Highlights

Semantic search + AI chat with inline citations
Smart runner pattern: one desktop keeps the index in sync, other devices can query
Built-in MCP server so external agents can query or fetch notes
Automatic deduplication, queue persistence, and rebuild workflows

Getting Started

Requirements

Google Gemini API key (get one free)
Obsidian desktop app for indexing (mobile can query/read-only)

Install

Option 1 – BRAT (recommended)

Install BRAT from Community Plugins.
BRAT settings → Add Beta Plugin → https://github.com/benbjurstrom/ezrag.
Enable EzRAG in Community Plugins.

Option 2 – Manual

Clone into your vault:

cd /path/to/vault/.obsidian/plugins
git clone https://github.com/benbjurstrom/ezrag

Build once:
```
cd ezrag
npm install
npm run build
```
Restart Obsidian and enable EzRAG.

Configure

Settings → EzRAG → enter your Gemini API key.
On desktop, toggle This machine is the runner to let it index.

Using EzRAG

Chat

Open via the ribbon icon or EzRAG: Open Chat. Try prompts like:

“What are my notes about the Johnson project?”
“Summarize yesterday’s meeting notes.”
“Find all mentions of machine learning.”

MCP Server

Enable Settings → EzRAG → MCP Server to let tools connect.

Connect from Claude Code:

claude mcp add --transport http ezrag-obsidian-notes http://localhost:42427/mcp

Tools provided:

keywordSearch – keyword/regex search
semanticSearch – Gemini-backed semantic search with citations
note:///<path> – direct note retrieval

How It Works

Indexing basics

Only .md files are indexed; changes trigger hashing + re-upload if content changed.
Runner enforcement prevents multiple machines from uploading the same file.
Upload queue persists across restarts and surfaces status in the UI.

Limits & costs

Gemini File Search pricing (details):

Indexing: ~$0.15 per 1M tokens (storage free; standard model rates for queries)
Max file size: 100 MB; free tier ≈1 GB total storage (higher tiers up to 1 TB)
For best performance keep stores under ~20 GB

Data control

Documents live in your Google account. Manage/delete stores via Settings → Manage Stores.
No telemetry or note data leaves your machine beyond the Gemini File Search uploads.