Anytype MCP Server

A Model Context Protocol (MCP) server for Anytype that enables semantic search and RAG (Retrieval-Augmented Generation) capabilities over your Anytype documents.

Features

• Semantic search across your Anytype documents
• Automatic chunking of documents for improved search accuracy
• Full metadata support including tags, dates, and custom properties

Prerequisites

• Python 3.13 or higher
• uv - Modern Python package installer
• Ollama - For running the embedding model locally
• Anytype desktop application

Installation

1. Clone the repository:

git clone https://github.com/wethegreenpeople/anytype-mcp.git
cd anytype-mcp

2. Install dependencies:

uv sync

3. Install the mxbai-embed-large model with Ollama:

ollama pull mxbai-embed-large

4. Optional: Run initial document ingestion (this can be done later from the client, but depending on your machine and how many notes you have, this might take a few minutes):

python scripts/ingest_only.py

5. Optional: Install the server in Claude:

fastmcp install server.py

Manual Configuration for Claude Desktop

If you want to manually add the Anytype MCP server to your Claude Desktop application, you'll need to modify the claude_desktop_config.json file. This file is typically located at:

• Windows: %APPDATA%\Claude\claude_desktop_config.json
• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

Add the following configuration to the mcpServers section of your claude_desktop_config.json:

"anytype": {
  "command": "uv",
  "args": [
    "run",
    "--with",
    "chromadb",
    "--with",
    "fastmcp",
    "--with",
    "httpx",
    "--with",
    "nltk",
    "--with",
    "ollama",
    "--with",
    "platformdirs",
    "--with",
    "sentence-transformers",
    "fastmcp",
    "run",
    "[PATH_TO_ANYTYPE_MCP]/server.py"
  ]
}

Replace [PATH_TO_ANYTYPE_MCP] with the absolute path to your anytype-mcp directory. For example:

• Windows: C:\\Users\\username\\Documents\\anytype-mcp\\server.py
• macOS/Linux: /home/username/anytype-mcp/server.py

Make sure to use double backslashes (\\) for Windows paths in the JSON file.

Authentication

The server handles authentication with Anytype via a challenge-based flow:

1. First run requires obtaining a challenge ID
2. You'll need to provide a secret code from Anytype
3. The server stores the authentication token for future sessions

Usage

The server provides several tools that can be used within Claude:

• ingest_documents: Ingest or update Anytype documents into the vector store
• query_anytype_documents: Perform semantic search across your documents
• get_anytype_object: Retrieve a specific Anytype object by ID
• get_ingestion_stats: View statistics about ingested documents
• clear_ingestion: Clear the vector store and start fresh

Technical Details

Embedding Model

The mxbai-embed-large model is used for generating embeddings in this project. It's fetched from Ollama and configured in the EmbeddingUtils class. If you prefer to use a different embedding model, you can modify the OLLAMA_MODEL variable in utils/embedding_utils.py.

Data Storage

The application uses platformdirs for cross-platform data storage:

• ChromaDB data: Stored in a platform-specific user data directory
• NLTK data: Downloaded and stored locally
• Authentication tokens: Securely stored in a configuration file

NLTK and Tokenization

This project uses the NLTK library for sentence tokenization, specifically the punkt_tab tokenizer. The tokenizer is automatically downloaded to the application's data directory during initialization.

License

Contributing