MCP Shrimp Task Manager

What cjo4m06/mcp-shrimp-task-manager MCP server does

cjo4m06/mcp-shrimp-task-manager MCP server organizes software-development work for AI coding agents. It supports project-rule initialization, task planning, decomposition into smaller units, dependency tracking, execution, progress review, research, and task reflection. Task history and progress are stored so an agent can continue work across sessions instead of rebuilding its context each time.

The workflows apply to feature development, bug fixing, and technical research. For example, an agent can plan authentication work, create a sequence of implementation tasks, execute those tasks, and review the result. Research mode can investigate alternatives before a follow-up implementation plan is created.

How it works

The server runs as a Node.js application after the repository is installed and built. An MCP client launches the compiled entry point and passes configuration through environment variables. The main persistent storage location is selected with DATA_DIR.

Typical interactions include commands such as init project rules, plan task, execute task, continuous mode, list tasks, research, and reflect task. Planning can split a larger request into atomic, testable work items, while dependency information determines relationships between tasks. Continuous mode processes tasks sequentially. Project rules provide a place to maintain coding standards for the project.

The task data can also be viewed through an optional lightweight web GUI. A separate React-based task viewer supports drag-and-drop, search, and multiple profiles according to its documentation.

Setup and configuration

cjo4m06/mcp-shrimp-task-manager MCP server requires Node.js 18 or newer and npm or yarn. Installation consists of cloning the repository, installing dependencies with npm install, and building with npm run build.

Configure the client to run node against the built dist/index.js file. Set DATA_DIR to the directory where task data should be kept. The examples also set TEMPLATES_USE to a language code and ENABLE_GUI to false; the configuration section documents ENABLE_GUI as optional. When the GUI is enabled, WEB_PORT can set a custom web port, and PROMPT_LANGUAGE can select a prompt language such as English or Chinese.

It also identifies Cline in VS Code and Claude Desktop as supported setup targets, with client-specific configuration locations provided for the latter.

Tools and capabilities

The documented command-oriented capabilities include:

• Initializing project rules and coding standards.
• Creating task plans from natural-language descriptions.
• Breaking complex work into smaller tasks.
• Tracking task dependencies and status.
• Executing a selected task or processing tasks continuously.
• Listing current tasks.
• Researching a topic before implementation.
• Reflecting on a task to review and improve it.
• Assigning specialized AI agents to particular tasks.
• Preserving task history through backup and restoration.

The optional task viewer adds a browser interface at the documented local address after its own dependencies are installed and its start command is run.

Limitations and notes

The project is designed around programming and AI-assisted development rather than general-purpose personal task management. The README requires a Node.js runtime, a package manager, and an MCP-compatible client. Persistent data depends on a correctly configured storage directory, so deployments should choose a location appropriate for the project and its backup needs.

The web GUI is optional, and the task viewer is a separate tool with its own installation steps. The README provides examples for Claude Code, Cline, and Claude Desktop; it does not document setup instructions for every MCP client. The project is released under the MIT License.