MCP Shrimp Task Manager
@cjo4m06
Shrimp Task Manager is a task tool built for AI Agents, emphasizing chain-of-thought, reflection, and style consistency. It converts natural language into structured dev tasks with dependency tracking and iterative refinement, enabling agent-like developer behavior in reasoning A
Overview
What is MCP Shrimp Task Manager?
MCP Shrimp Task Manager is an MCP (Model Context Protocol) server that provides intelligent task management for AI-powered development. It helps break down complex projects into manageable tasks, maintains context across sessions, and is designed for developers using MCP-compatible AI clients like Claude Code.
How to use MCP Shrimp Task Manager?
Install Node.js 18+, clone the repository, then run npm install and npm run build. Configure the server by creating a .mcp.json file with the path to the built dist/index.js and setting the DATA_DIR environment variable. Start your AI client with the custom MCP config, then use commands like "init project rules", "plan task", "execute task", or "continuous mode".
Key features of MCP Shrimp Task Manager
- Persistent memory across AI sessions
- Structured workflows for planning and execution
- Smart decomposition of complex tasks
- Automatic dependency tracking between tasks
- Research mode for systematic exploration
- Optional web GUI and task viewer
Use cases of MCP Shrimp Task Manager
- Feature development: plan and implement new functionality step by step
- Bug fixing: research issues and execute fix tasks in continuous mode
- Research and learning: systematically compare technologies or approaches
FAQ from MCP Shrimp Task Manager
What are the runtime requirements?
Node.js 18+ with npm or yarn, and an MCP-compatible AI client such as Claude Code.
Where does task data live?
Task data is stored in the directory specified by the DATA_DIR environment variable in your configuration.
What transport does the server use?
The server communicates via MCP stdio transport, launched with a node command pointing to dist/index.js.
Is authentication supported?
No authentication mechanisms are mentioned in the documentation.
Is there a graphical interface?
Yes, an optional web GUI can be enabled by setting ENABLE_GUI=true in the configuration, and a separate React-based Task Viewer is available in the tools/task-viewer directory.
