mcp-task-manager-server

MCP.Pizza Chef: bsmi021

The MCP Task Manager Server is a local backend server compliant with the Model Context Protocol, designed to manage structured project and task data persistently using a SQLite database. It provides standardized tools for MCP clients to interact with task data organized by projects, enabling client-driven workflows without imposing specific logic. This server is ideal for AI agents or scripts requiring reliable, local task management within MCP ecosystems.

Use This MCP server To

Store and retrieve project-specific task data locally Enable AI agents to manage tasks within defined projects Persist task information using a lightweight SQLite database Provide standardized task management tools for MCP clients Support client-driven workflows for project and task handling Integrate with local AI scripts for task automation Maintain organized task data for multi-project environments

README

MCP Task Manager Server

MCP Task Manager Logo

A local Model Context Protocol (MCP) server providing backend tools for client-driven project and task management using a SQLite database.

Overview

This server acts as a persistent backend for local MCP clients (like AI agents or scripts) that need to manage structured task data within distinct projects. It handles data storage and provides a standardized set of tools for interaction, while the strategic workflow logic resides within the client.

Key Features:

  • Project-Based: Tasks are organized within distinct projects.
  • SQLite Persistence: Uses a local SQLite file (./data/taskmanager.db by default) for simple, self-contained data storage.
  • Client-Driven: Provides tools for clients; does not dictate workflow.
  • MCP Compliant: Adheres to the Model Context Protocol for tool definition and communication.
  • Task Management: Supports creating projects, adding tasks, listing/showing tasks, updating status, expanding tasks into subtasks, and identifying the next actionable task.
  • Import/Export: Allows exporting project data to JSON and importing from JSON to create new projects.

Implemented MCP Tools

The following tools are available for MCP clients:

  • createProject:
    • Description: Creates a new, empty project.
    • Params: projectName (string, optional, max 255)
    • Returns: { project_id: string }
  • addTask:
    • Description: Adds a new task to a project.
    • Params: project_id (string, required, UUID), description (string, required, 1-1024), dependencies (string[], optional, max 50), priority (enum 'high'|'medium'|'low', optional, default 'medium'), status (enum 'todo'|'in-progress'|'review'|'done', optional, default 'todo')
    • Returns: Full TaskData object of the created task.
  • listTasks:
    • Description: Lists tasks for a project, with optional filtering and subtask inclusion.
    • Params: project_id (string, required, UUID), status (enum 'todo'|'in-progress'|'review'|'done', optional), include_subtasks (boolean, optional, default false)
    • Returns: Array of TaskData or StructuredTaskData objects.
  • showTask:
    • Description: Retrieves full details for a specific task, including dependencies and direct subtasks.
    • Params: project_id (string, required, UUID), task_id (string, required)
    • Returns: FullTaskData object.
  • setTaskStatus:
    • Description: Updates the status of one or more tasks.
    • Params: project_id (string, required, UUID), task_ids (string[], required, 1-100), status (enum 'todo'|'in-progress'|'review'|'done', required)
    • Returns: { success: true, updated_count: number }
  • expandTask:
    • Description: Breaks a parent task into subtasks, optionally replacing existing ones.
    • Params: project_id (string, required, UUID), task_id (string, required), subtask_descriptions (string[], required, 1-20, each 1-512), force (boolean, optional, default false)
    • Returns: Updated parent FullTaskData object including new subtasks.
  • getNextTask:
    • Description: Identifies the next actionable task based on status ('todo'), dependencies ('done'), priority, and creation date.
    • Params: project_id (string, required, UUID)
    • Returns: FullTaskData object of the next task, or null if none are ready.
  • exportProject:
    • Description: Exports complete project data as a JSON string.
    • Params: project_id (string, required, UUID), format (enum 'json', optional, default 'json')
    • Returns: JSON string representing the project.
  • importProject:
    • Description: Creates a new project from an exported JSON string.
    • Params: project_data (string, required, JSON), new_project_name (string, optional, max 255)
    • Returns: { project_id: string } of the newly created project.
  • updateTask:
    • Description: Updates specific details (description, priority, dependencies) of an existing task.
    • Params: project_id (string, required, UUID), task_id (string, required, UUID), description (string, optional, 1-1024), priority (enum 'high'|'medium'|'low', optional), dependencies (string[], optional, max 50, replaces existing)
    • Returns: Updated FullTaskData object.
  • deleteTask:
    • Description: Deletes one or more tasks (and their subtasks/dependency links via cascade).
    • Params: project_id (string, required, UUID), task_ids (string[], required, 1-100)
    • Returns: { success: true, deleted_count: number }
  • deleteProject:
    • Description: Permanently deletes a project and ALL associated data. Use with caution!
    • Params: project_id (string, required, UUID)
    • Returns: { success: true }

(Note: Refer to the corresponding src/tools/*Params.ts files for detailed Zod schemas and parameter descriptions.)

Getting Started

  1. Prerequisites: Node.js (LTS recommended), npm.

  2. Install Dependencies:

    npm install
  3. Run in Development Mode: (Uses ts-node and nodemon for auto-reloading)

    npm run dev

    The server will connect via stdio. Logs (JSON format) will be printed to stderr. The SQLite database will be created/updated in ./data/taskmanager.db.

  4. Build for Production:

    npm run build
  5. Run Production Build:

    npm start

Configuration

  • Database Path: The location of the SQLite database file can be overridden by setting the DATABASE_PATH environment variable. The default is ./data/taskmanager.db.
  • Log Level: The logging level can be set using the LOG_LEVEL environment variable (e.g., debug, info, warn, error). The default is info.

Project Structure

  • /src: Source code.
    • /config: Configuration management.
    • /db: Database manager and schema (schema.sql).
    • /repositories: Data access layer (SQLite interaction).
    • /services: Core business logic.
    • /tools: MCP tool definitions (*Params.ts) and implementation (*Tool.ts).
    • /types: Shared TypeScript interfaces (currently minimal, mostly in repos/services).
    • /utils: Logging, custom errors, etc.
    • createServer.ts: Server instance creation.
    • server.ts: Main application entry point.
  • /dist: Compiled JavaScript output.
  • /docs: Project documentation (PRD, Feature Specs, RFC).
  • /data: Default location for the SQLite database file (created automatically).
  • tasks.md: Manual task tracking file for development.
  • Config files (package.json, tsconfig.json, .eslintrc.json, etc.)

Linting and Formatting

  • Lint: npm run lint
  • Format: npm run format

(Code is automatically linted/formatted on commit via Husky/lint-staged).

mcp-task-manager-server FAQ

How does the MCP Task Manager Server store data?
It uses a local SQLite database file (default: ./data/taskmanager.db) for persistent storage.
Can I customize the workflow logic for task management?
Yes, the server provides tools but leaves workflow logic to the client, allowing full customization.
Is the MCP Task Manager Server suitable for remote or cloud deployment?
It is primarily designed as a local server but can be adapted for remote use with appropriate configuration.
How does the server ensure compliance with the Model Context Protocol?
It adheres to MCP standards for tool definition and communication, ensuring seamless integration with MCP clients.
What kind of clients can interact with this server?
Any MCP-compliant client, including AI agents and scripts, can use the provided tools to manage tasks.
Does the server support multiple projects simultaneously?
Yes, tasks are organized within distinct projects to support multi-project management.
How do I set up the MCP Task Manager Server?
Clone the repository, ensure SQLite is installed, and run the server; it uses a local SQLite file by default.
Can the database location be changed?
Yes, the SQLite database file path can be configured as needed.