Claude Task Master

Definition

Task Master is a task management system for AI-driven development that works in Cursor, Claude Code, Windsurf, Roo, Lovable, and other AI chats. It parses a PRD into tasks, auto-expands them into subtasks, tracks progress, and integrates with many AI providers. It runs as an MCP server, so the same task workflow reaches every supported editor. The code lives in a Turborepo monorepo with separate packages for the MCP server, the CLI, and the frontend apps. Configuration sits in a small JSON file beside your editor settings. Every task carries an ID, title, description, status, dependencies, priority, subtasks, complexity score, and tags. A tag system keeps workstreams apart, and tasks move between tags with one command.

Basic Information

Installation

Claude Code

claude mcp add taskmaster-ai -- npx -y task-master-ai

Cursor (One-Click)

Add task-master-ai MCP server to Cursor

NPM

npm install -g task-master-ai
task-master init

Architecture

claude-task-master/
├── .claude-plugin/   # Plugin manifests
├── .taskmaster/      # Task data & configs
├── apps/             # Frontend apps
├── context/          # Context templates
├── docs/             # Documentation
├── mcp-server/       # MCP server implementation
├── packages/         # Core packages
├── scripts/          # Utility scripts
├── src/              # Core source
└── tests/            # Test suite

Core Features

PRD to Tasks

Parses a Product Requirements Document into structured tasks with dependencies, priorities, and complexity scores.

Task Structure

Each task has:

• ID, title, description
• Status (pending, in-progress, done, etc.)
• Dependencies (task IDs)
• Priority (high, medium, low)
• Subtasks
• Complexity score
• Tags

Multi-AI Provider Support

MCP Integration

36 MCP tools available (~21,000 tokens):

AI Research

task-master research "What are the latest best practices for JWT authentication?"

Researches with project context and cross-references findings.

Task Commands

Cross-Tag Movement

task-master move --from=5 --from-tag=backlog --to-tag=in-progress
task-master move --from=5,6,7 --from-tag=backlog --to-tag=done --with-dependencies

Configuration

MCP config in .cursor/mcp.json, .vscode/mcp.json, etc.:

{
  "mcpServers": {
    "task-master-ai": {
      "command": "npx",
      "args": ["-y", "task-master-ai"],
      "env": {
        "ANTHROPIC_API_KEY": "...",
        "PERPLEXITY_API_KEY": "...",
        "TASK_MASTER_TOOLS": "standard"
      }
    }
  }
}

Tool Loading Configuration

# Claude Code CLI
claude mcp add task-master-ai --scope user \
  --env TASK_MASTER_TOOLS="core" \
  -- npx -y task-master-ai@latest

Tech Stack

Pros

Cons

When to Use

• New projects: Start with a PRD and generate structured tasks
• Multi-editor teams: Team members use different editors
• Task-driven development: Want structured task management
• AI-assisted planning: Need AI to expand and research tasks
• MCP-first workflows: Want task management through MCP tools

The install path is short. Claude Code adds the server with one command. Cursor offers a one-click MCP link. A global npm install finishes with task-master init. Turborepo holds the MCP server, the CLI, the apps, and the tests. Parsing a PRD builds tasks with status, priority, dependencies, and a complexity score. You can list them, ask for the next one, expand a task, or show a few IDs. Research uses project context. Providers include Anthropic, OpenAI, Google, Perplexity, xAI, and OpenRouter. Claude Code needs no API key, and Codex CLI uses ChatGPT OAuth. The default loads 36 tools at about 21k tokens. Standard keeps 15 near 10k, and core keeps 7 near 5k. Move can send tasks to another tag with their dependencies. The license is MIT plus a Commons Clause, so you cannot sell the product as a service. Most other models still need a key.

References:

• eyaltoledano/claude-task-master
• npm: task-master-ai
• Docs

See also: Kit Comparison · Other Kits · Coding Agents