LLM-Driven Productivity Tracker with MCP Integration

An intelligent task management system powered by LangChain agents and Model Context Protocol (MCP), enabling natural language interaction for productivity tracking.

Features

• Multi-step AI Agent: Autonomous decision-making with LangChain for complex task workflows

• Natural Language Interface: Interact with your tasks using conversational AI

• MCP Server Integration: Exposes productivity tools via standardized protocol

• Claude Desktop Compatible: Works seamlessly as an MCP client

• Comprehensive Task Management:

  • Log tasks with 8 categories (work, health, learning, personal, finance, social, hobby, self_care)

  • 4 status types (todo, started, completed, blocked)

  • Update and remove tasks by ID or name

  • Time-based summaries (daily, weekly, monthly)

• Local LLM Support: Runs on Ollama with llama3.2 (no API costs)

Technologies

• Python 3.12

• LangChain (Agent framework)

• Ollama (llama3.2)

• Model Context Protocol (MCP)

• SQLite (Database)

• Claude Desktop API

• JSON-RPC

Prerequisites

• Python 3.12+

• Ollama installed with llama3.2 model

• Claude Desktop (optional, for MCP integration)

Installation

1. Clone the repository: git clone cd ProductivityTracker

2. Install dependencies: pip install -r requirements.txt

3. Install Ollama and pull llama3.2: ollama pull llama3.2

4. Initialize the database: python -c "import database; database.init_db()"

Usage

Option 1: Local Agent (agent.py)

Run the agent locally with your own questions:

python agent.py

Modify the question variable in agent.py to test different queries.

Option 2: MCP Server with Claude Desktop

1. Configure Claude Desktop by editing %APPDATA%\Claude\claude_desktop_config.json:

{ "mcpServers": { "productivity-tracker": { "command": "C:\path\to\python.exe", "args": ["C:\path\to\ProductivityTracker\mcp_server.py"] } } }

2. Restart Claude Desktop

3. Interact naturally:

   • "Log a task to review code for work as started"

   • "Show me today's summary"

   • "Update test task to completed"

Project Structure

ProductivityTracker/ ├── agent.py # Local LangChain agent with multi-step reasoning ├── mcp_server.py # MCP server exposing tools via protocol ├── tools.py # Tool definitions (log_task, get_summary, etc.) ├── database.py # SQLite operations with error handling ├── requirements.txt # Python dependencies └── productivity_tracker.db # SQLite database (auto-created)

Example Interactions

Log a task: "Log a morning workout for health category as completed"

Get summary: "How was my week?"

Update task: "Mark the code review task as completed"

Remove task: "Delete the duplicate dinner task"

Architecture

• Agent Pattern: AI decides which tools to use based on user intent

• Tool Pattern: 5 custom tools for task management operations

• MCP Integration: Standardized protocol for external agent communication

• Iterative Loop: Multi-step reasoning with context maintenance

Error Handling

• Database connection errors handled gracefully

• Invalid category/status inputs caught with helpful messages