Taskvault AI MCP

TaskVault-AI-MCP

Task Vault MCP Server is a project designed to expose task management capabilities via an MCP server. It allows you to effortlessly create, store, and manage your tasks using just natural language. Whether you want to add a new to-do, check your agenda for the day, or look up details about a specific project, simply ask any MCP compatible LLM client. All your tasks are securely stored locally and easily retrievable, making personal organization as simple as a conversation.

Features

• Client Agnostic MCP Tools:

Exposes a set of Model Context Protocol (MCP) tools for any client to use. Can be used by any MCP-compliant client to manage tasks.

• ChromaDB Storage:

Uses ChromaDB to store tasks locally and reliably. This makes the tasks accessible between chat sessions and across different llm applications

How It Works

1. Storage: The server uses ChromaDB to maintain the state of tasks.
2. Access: Clients connect to the server and utilize the exposed tools.
3. Management: Users can add, retrieve, and manage tasks through the MCP server interface to the LLM/MCP client.

Installation

Prerequisites

• Python3.12 or above
• uv package manager
• Ollama
• An MCP client (e.g., using Claude Desktop). The project also includes a simple cli based MCP client for quick setup.

Steps to connect to the MCP server using Claude Desktop

[!NOTE] This requires paid plan of Claude Desktop.

1. Clone/download the repository
2. Download and install ollama from
3. Download embedding model mxbai-embed-large by running

4. Install Claude Desktop from 
5. Go to developer settings in using Claude Desktop. 

(Settings -> Desktop App -> Developer -> Local MCP servers -> Edit config)
![Claude Desktop mcp connector settings](docs/claude_desktop_mcp_server_setting.png)
6. This opens folder with claude_desktop_config.json. Open this file and Paste the below code. Please change the absolute path to the repository.

{ "mcpServers": { "task_storage_service": { "command": "uv", "args": [ "--directory", "//TaskVault-AI-MCP", "run", "task_storage_service" ] } } }

1. Restart Claude Desktop and start a new chat. the MCP server "task_storage_service" should be listed under Connectors.

Steps to connect to the MCP server using using Ollama and inbuilt MCP client.

[!NOTE] This is free but requires sufficient RAM to run small llm models like IBM's granite4.

1. Clone/download the repository
2. Download and install ollama from
3. Download llm model granite4:3b-h and embedding model mxbai-embed-large by running

4. Run in cmd ```uv tool install . -e```.
 This will automatically install all depedencies and create an executable taskvault.exe.
5. Run in cmd  ```taskvault```

![Sample Chat using inbuilt mpc client](docs/Inbuilt_mcp_client_sample_chat.png)