daz-obsidian-mcp

#  # daz obsidian mcp **A powerful toolkit for indexing, searching, and serving Obsidian notes with natural language queries via ChromaDB and Model Context Protocol (MCP).** --- ## 🚀 Features - **Automatic Indexing**: Watches your Obsidian vault and indexes Markdown files into overlapping content chunks for fast semantic search. - **Vector Search**: Leverages ChromaDB to find relevant snippets or entire note contents for any query. - **MCP Server**: Exposes search tools via the Model Context Protocol, compatible with AI-powered workflows. - **Flexible Setup**: Supports vault autodiscovery or configuration via environment variables. - **Live Updates**: Near real-time updates as your vault changes. --- ## 📁 Project Structure - `scanner.py` — File watcher and content indexer for your Obsidian vault - `daz-obsidian-mcp.py` — MCP server exposing semantic search tools (snippets/full article) - `search.py` — Simple CLI search/test utility - `obsidian_chromadb/` — Directory where the vector store (ChromaDB) is kept --- ## 🛠️ Setup ### 1. Requirements - Python 3.8+ - pip (`pip install -r requirements.txt` — see below) **Dependencies:** - `chromadb` - `watchdog` - `mcp` (Model Context Protocol Python SDK) Install with: ```bash pip install chromadb watchdog mcp ``` --- ### 2. Connect Your Obsidian Vault By default, the indexer will try to auto-discover your vault (macOS conventions supported): - `~/Documents/Obsidian/` or `~/Library/Mobile Documents/iCloud~md~obsidian/Documents/` **Custom Vault:** Set the environment variable before running: ```bash export OBSIDIAN_VAULT_PATH=/path/to/your/ObsidianVault ``` --- ## 🔎 Usage ### 1. Index Your Vault (Run the Watcher) ```bash python scanner.py ``` This script creates/updates the vector database in `obsidian_chromadb/`, performing full and incremental indexing as your .md files change. ### 2. Test Search Manually (Optional) ```bash python search.py ``` - Check chunk counts - Run basic test queries - Preview sample indexed files ### 3. Connect the MCP Search Server in Your AI App/Client Instead of running the server manually, add the following command to your Claude Desktop, ChatGPT, or other MCP-compatible client as a tool/plugin configuration: ``` python daz-obsidian-mcp.py ``` This lets your chosen client/tool launch and interact with the server directly. See your platform's documentation for how to add a custom tool/command. --- ## 🧑‍💻 Example MCP Tool Usage #### Query for Snippets: ```json { "tool": "search_snippets", "arguments": { "query": "What is spaced repetition?", "limit": 5 } } ``` #### Query for Full Note Content: ```json { "tool": "search_full", "arguments": { "query": "zettelkasten", "limit": 2 } } ``` --- ## 📝 Notes & Troubleshooting - If you see "ChromaDB database not found", run `scanner.py` first. - To re-index everything: delete the `obsidian_chromadb/` folder and metadata, rerun the indexer. - Chunking is overlapping; deduplication for full-text is handled simply by concatenation. - UTF-8 Safety: Chunks are tokenized on byte-level but also respect character boundaries to avoid decode errors. --- ## 🗂️ License MIT License (c) 2024 — Your Name Here --- ## 🙏 Acknowledgements - [ChromaDB](https://www.trychroma.com/) - [Obsidian](https://obsidian.md/) - [MCP](https://github.com/modelcontext/protocol) - [Watchdog](https://python-watchdog.readthedocs.io/) --- > ⬆️ _Add your own **banner.jpg** image at the top of this folder to complete the visual branding!_