Task Master 🤖 Agent Free

What it does

Task Master is a task-management layer for AI-driven development. You give it a product requirements document, it breaks the document into a list of tasks with dependencies, priorities and subtasks, and your coding agent then works through them one at a time, asking what comes next rather than trying to build everything in one pass. It runs as an MCP server inside your editor or as a standalone command-line tool. Its documentation now lives on the Hamster website.

What is inside / how it works

Tasks are stored in a .taskmaster/ folder in your project. Core commands parse a PRD, list tasks, show the next one, expand a task into subtasks, analyse complexity and move work between tags that act as separate workstreams. A research command pulls in fresh information with project context. Task Master calls a language model for these steps, and you can configure a main, research and fallback model from providers including Anthropic, OpenAI, Google, Perplexity, xAI and OpenRouter; Claude Code and Codex CLI can also be used as providers without a separate API key. The MCP server exposes up to 36 tools, and a TASK_MASTER_TOOLS setting loads a smaller core or standard set to save context.

Works with

The supported-editors page lists Cursor, Claude Code (including a plugin with slash commands and agents), Windsurf, VS Code, Roo, Cline, Codex, Gemini CLI, Kiro, Zed, Amp, Trae and OpenCode.

How to install or start

For Claude Code, add the MCP server, then put your API keys in the project's .env or the MCP config:

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

For the command line, install it and initialise a project:

npm install -g task-master-ai
task-master init
task-master parse-prd your-prd.txt

Pricing

The tool itself is free, but it uses the MIT licence with a Commons Clause, which forbids selling Task Master or offering it as a hosted service. You pay for any model API calls it makes. Hamster's separate hosted products have their own pricing on the official site.

Who should use it

Developers building features large enough to plan, who find agents lose track midway. Results depend heavily on the quality of your PRD, and the full MCP tool set uses a noticeable share of the context window.