HarnessCommunity★ 27k

Claude Task Master

Updated Jul 7, 2026

An AI task-management system (now called "Taskmaster", maintained by eyaltoledano and RalphEcom) that parses a PRD into structured tasks with dependency graphs and complexity scores, then exposes them to Claude Code and other editors through an MCP server. Supports Anthropic, OpenAI, Gemini, Perplexity, xAI, OpenRouter, Azure OpenAI, Groq, Ollama, and Claude Code CLI (no API key required).

View on GitHub

What it does

parse_prd
Parses a PRD document and generates a structured task list with dependency graph and complexity scores.
next_task
Returns the next unblocked task with full context so Claude Code implements tasks in dependency order.
expand_task
Breaks a high-complexity task into sub-tasks with their own dependency annotations.
analyze_project_complexity
Scores all tasks by complexity and outputs a report to guide sub-task expansion priorities.
research
Runs a web-augmented, fresh-context research pass to provide current documentation for a task before implementation.
task-master init
Initializes Taskmaster in a project, creating the .taskmaster/ directory with config and example PRD template.

Files (1)

OVERVIEW.mdprimary · markdown · 1.8 KB

## What it does

Taskmaster (github.com/eyaltoledano/claude-task-master) is an AI task-management system that parses a PRD into structured tasks with dependency graphs and complexity scores, then feeds them one at a time to Claude Code through an MCP server. Claude Code calls next_task to get the next unblocked task with full context, implements it, marks it done, and repeats — always in dependency order.

Supports 36 MCP tools (configurable via TASK_MASTER_TOOLS env var), a CLI, and multiple AI providers for the main, research, and fallback models. Selectable tool loading reduces context window usage: "core" mode (7 tools) uses ~5K tokens vs "all" (36 tools) ~21K tokens.

Key MCP tools: parse_prd, next_task, get_task, set_task_status, expand_task, analyze_project_complexity, add_task, add_subtask, complexity_report, research (web-augmented fresh-context research), and move (cross-tag task movement).

Tags and workstreams let you organize tasks across parallel tracks (e.g., backlog, in-progress, done) with cross-tag movement.

## Install

Claude Code (via MCP):

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

Add API keys to the env section of the MCP config or your project .env. Minimum: one of ANTHROPIC_API_KEY, OPENAI_API_KEY, GOOGLE_API_KEY, or use Claude Code CLI with no key via claude-code/sonnet or claude-code/opus.

Initialize:

task-master init
# then in your AI chat: "Initialize taskmaster-ai in my project"

Parse a PRD:

task-master parse-prd your-prd.txt

## When to use

Works well for projects driven from a written PRD where you want AI to respect task dependencies automatically. The complexity scorer helps prioritize which tasks need more detailed sub-task expansion before implementation. The research tool augments task context with fresh web information when working with recent APIs or rapidly-changing libraries.