obsidian-semantic-search

Semantic search for Obsidian vaults using AWS Bedrock embeddings and LanceDB. Search via Claude MCP, sync via minimal Obsidian plugin.

How it works

Obsidian Vault
      │
      ▼
Sync Plugin ──► Python Backend ──► AWS Bedrock (embeddings)
                     │
                     ▼
                  LanceDB (local vector store)
                     │
                     ▼
               MCP Server ──► Claude

Search: Ask Claude to find notes by meaning, not just keywords.
Sync: Trigger indexing from Obsidian via command palette.

Components

Component	Description
backend/	Python: embeddings, indexing, MCP server, HTTP API
plugin/	Obsidian plugin: sync triggers, status bar

Requirements

• Python 3.11+
• AWS account with Bedrock access (Titan Embeddings V2)
• Obsidian 1.0+

Quick Start

1. Install backend

cd backend
pip install -e .

2. Configure

Create ~/.config/obsidian-semantic-search/config.toml:

[vault]
path = "/path/to/your/vault"

[bedrock]
region = "eu-west-1"

3. Initial index

obsidian-semantic-search reindex --full

4. Start the backend

# HTTP API for Obsidian plugin
uvicorn obsidian_semantic_search.api:app --port 8765

5. Configure Claude

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "obsidian-semantic": {
      "command": "python",
      "args": ["-m", "obsidian_semantic_search.server"]
    }
  }
}

6. Install Obsidian plugin

Copy plugin/ contents to .obsidian/plugins/semantic-sync/ and enable in settings.

MCP Tools

Tool	Description
semantic_search	Find notes by meaning
similar_notes	Find notes similar to a given note
reindex	Trigger full or incremental reindex

Cost

Bedrock Titan Embeddings: ~$0.0001 per 1K tokens.
A 500-note vault costs roughly $0.05 to index initially, under €1/year ongoing.

License

MIT