Claude Task Master

What eyaltoledano/claude-task-master MCP server does

eyaltoledano/claude-task-master MCP server provides task planning and management for AI-assisted software development. Its workflow starts with a product requirements document (PRD), which can be parsed into tasks and then expanded into more detailed implementation steps. Tasks support dependencies, tags, and workstreams, giving an AI client more structure than an untracked list of prompts.

The project is designed for use through MCP-compatible development environments. The README specifically documents Cursor, Windsurf, VS Code, and Q Developer CLI configuration, and describes the system as working with Claude and Cursor. It also includes CLI commands alongside its MCP interface.

How it works

The server runs as a local standard-input/standard-output process using the task-master-ai npm package. An MCP client starts it with npx -y task-master-ai, then communicates with the process using the client’s MCP configuration. The server uses configured AI models for operations such as task generation, task expansion, and research.

Three model roles are supported: a main model, a research model, and a fallback model. The research model is optional. Provider selection can include Anthropic, OpenAI, Google Gemini, Perplexity, xAI, OpenRouter, and other providers shown in the configuration example. Claude Code and Codex CLI are also listed as alternatives to directly supplying an API key, with Codex CLI using OAuth through a ChatGPT subscription.

Setup and configuration

Install the package through npm and register it in the MCP client configuration. Cursor and Windsurf use an mcpServers object, while VS Code uses a servers object and includes a server type in its configuration format. The README gives separate configuration locations for global and project-level setups.

At least one supported provider credential is needed for the AI-backed commands unless Claude Code or Codex CLI is being used. Keys can be placed in the project’s root .env file or in the MCP configuration’s env section. The one-click Cursor setup uses placeholder values, so those values must be replaced before use. If the client reports that no tools are enabled, the documented checks are to restart the editor and verify the API key configuration.

Tools and capabilities

eyaltoledano/claude-task-master MCP server supports capabilities documented around:

• Parsing PRDs into tasks
• Expanding tasks into implementation details
• Managing task status and dependencies
• Organizing work with tags and workstreams
• Running research through a selected research model
• Automating workflows with a loop command
• Loading all, standard, core, or selected tools through TASK_MASTER_TOOLS

The available tool set can therefore be narrowed to a comma-separated selection instead of exposing every capability to the client.

Limitations and notes

The server does not remove the need for model-provider access. Most AI operations require at least one applicable API key, and using several providers requires configuring their respective keys. Research is optional, but it uses a separate research-model configuration when enabled.

The provided material documents Cursor, Windsurf, VS Code, and Q Developer CLI setup, but does not establish compatibility with every MCP client. The package is open source under an MIT license with a Commons Clause, as indicated by the repository badge; the Commons Clause may add restrictions beyond a standard MIT license.