Taskmaster (also written Task Master, and published on npm as task-master-ai) is a task-management
system for AI-driven development. Its README describes it as designed to work with any AI chat, and
in particular as built for development with Claude alongside Cursor; the
repository’s description adds that it can be dropped into Cursor, Lovable, Windsurf, Roo and other
tools. It turns a project’s requirements into a structured list of tasks that the developer’s AI
assistant then plans against, breaks down and implements one at a time. Its documentation is published
on the website of Hamster, whose other products the README also links to.
Capabilities
The workflow the README recommends starts from a product requirements document: “Always start with a detailed PRD”, on the stated reasoning that the more detailed the PRD, the better the generated tasks. Taskmaster parses the PRD into tasks, and the developer then asks for the next task to work on, for help implementing a particular task, or for a task to be expanded into subtasks; individual tasks can also be created directly from a request in chat without a PRD. Tasks carry dependencies and can be organised under tags, with commands to move tasks between tags with or without their dependencies. A research command fetches fresh information, optionally with project context, and the documentation also covers analysing task complexity, team collaboration and a loop command for automation.
Several of those commands call an LLM, so Taskmaster needs at least one provider configured. It distinguishes three model roles — a main model, a research model, and a fallback model used when either of the others fails — and supports Anthropic, OpenAI, Google Gemini, Perplexity (suggested for research), xAI, OpenRouter and others through API keys, as well as Claude Code and the Codex CLI (OpenAI Codex) through their own sign-in, with no API key required.
There are two ways to run it. The recommended one is as a Model Context Protocol
server configured in the editor — the README gives configuration paths for Cursor, Windsurf, VS Code
and Amazon Q Developer CLI, and a one-line install for Claude Code — after which the developer drives
it by talking to the editor’s AI chat (“Initialize taskmaster-ai in my project”, “What’s the next task
I should work on?”). The other is the task-master command-line tool, with commands such as
parse-prd, list, next, show and research, and an init step that can also install rule files
for particular editors.
Because a large tool list costs context, the MCP server supports selective tool loading. By default
all 36 of its tools are loaded, which the README puts at roughly 21,000 tokens; a standard set of 15
tools and a core set of 7 (for example getting tasks, finding the next task, setting a task’s status,
parsing a PRD and expanding a task) reduce that, and a custom comma-separated list is also accepted.
The README recommends the standard set for new users and the core set for large projects.
Adoption & Ecosystem
Taskmaster is licensed under the MIT License with the Commons Clause. As the README summarises it, the tool may be used for any purpose, modified and redistributed, and products built with it may be sold, but Taskmaster itself may not be sold, offered as a hosted service, or used as the basis of a competing product.