English | 中文

• ✨ Features
• 🧭 Usage Guide
• 🔬 Research Mode
• 🧠 Task Memory Function
• 📋 Project Rules Initialization
• 🌐 Web GUI
• 📚 Documentation Resources
• 🔧 Installation and Usage
• 🧪 Testing and Development
• 🔌 Using with MCP-Compatible Clients
• 💡 System Prompt Guidance
• 🛠️ Available Tools Overview
• 📄 License
• 🤖 Recommended Models

🚀 An intelligent task management system based on Model Context Protocol (MCP), providing an efficient programming workflow framework for AI Agents.

Shrimp Task Manager guides Agents through structured workflows for systematic programming, enhancing task memory management mechanisms, and effectively avoiding redundant and repetitive coding work.

• 🎯 PRD-Driven Task Generation: Intelligently generate role-specific professional tasks based on Product Requirements Documents (PRD)
• 👥 Role-Specialized Task Templates: Professional task templates for Product Managers, Designers, Developers, QA Engineers, and more
• 🤝 Cross-Role Collaboration Detection: Automatically identify team collaboration needs and generate handoff points
• 📊 Intelligent Workflow Management: Generate complete project workflows, critical paths, and parallel execution tracks
• 🔗 Smart Dependency Management: Automatically analyze and validate task dependencies, preventing circular dependencies

• Task Planning and Analysis: Deep understanding and analysis of complex task requirements
• Intelligent Task Decomposition: Automatically break down large tasks into manageable smaller tasks
• Dependency Management: Precisely handle dependencies between tasks, ensuring correct execution order
• Execution Status Tracking: Real-time monitoring of task execution progress and status
• Task Completeness Verification: Ensure task results meet expected requirements
• Task Complexity Assessment: Automatically evaluate task complexity and provide optimal handling suggestions
• Automatic Task Summary Updates: Automatically generate summaries upon task completion, optimizing memory performance

• Task Memory Function: Automatically backup task history, providing long-term memory and reference capabilities
• 🧠 Dynamic Task Memory & Context Awareness: Intelligent task insertion and context-aware adjustments during execution
• 💡 Team Knowledge Sharing: Automatic knowledge accumulation and sharing across team members
• 🔄 Smart Task Adjustment: Dynamically adjust subsequent tasks based on execution discoveries and decisions
• 📚 Execution History & Pattern Recognition: Learn from past executions to improve future task planning
• 🤝 Advanced Team Collaboration Memory (NEW!): Complete team memory system with knowledge sharing, collaboration pattern learning, and collective intelligence
• 🔄 Dynamic Task Insertion (NEW!): Intelligent task insertion with context awareness and dependency resolution
• 📊 Team Performance Analytics (NEW!): Analyze team collaboration effectiveness and identify improvement opportunities
• Research Mode: Systematic technical research capabilities with guided workflows for exploring technologies, best practices, and solution comparisons
• Project Rules Initialization: Define project standards and rules to maintain consistency across large projects

• Web GUI: Provides an optional web-based graphical user interface for task management. Enable by setting ENABLE_GUI=true in your .env file. When enabled, a WebGUI.md file containing the access address will be created in your DATA_DIR
• 📁 Documentation Path Management: Structured document organization with project-specific paths for analysis, reflection, research, and team collaboration documents
• Multi-Project Support: Automatic project detection with isolated data directories for different projects
• Real-time Data Isolation: Each project's task data is completely isolated to avoid conflicts

Shrimp Task Manager offers a structured approach to AI-assisted programming through guided workflows and systematic task management.

Shrimp is essentially a prompt template that guides AI Agents to better understand and work with your project. It uses a series of prompts to ensure the Agent aligns closely with your project's specific needs and conventions.

Before diving into task planning, you can leverage the research mode for technical investigation and knowledge gathering. This is particularly useful when:

• You need to explore new technologies or frameworks
• You want to compare different solution approaches
• You're investigating best practices for your project
• You need to understand complex technical concepts

Simply tell the Agent "research [your topic]" or "enter research mode for [technology/problem]" to begin systematic investigation. The research findings will then inform your subsequent task planning and development decisions.

Our core differentiator is the ability to generate professional, role-specific tasks directly from Product Requirements Documents (PRD). This feature is designed for team collaboration and addresses the gap in existing task management tools.

How to use:

• Tell the Agent: "Generate team tasks from this PRD: [paste your PRD content]"
• Or use the tool directly: generate_team_collaboration_tasks with your PRD content

What makes us different:

• Role-Specialized Tasks: Automatically generates tasks tailored for Product Managers, UI/UX Designers, Frontend/Backend Developers, QA Engineers, DevOps Engineers, and more
• Cross-Role Collaboration: Identifies collaboration points and handoff requirements between different roles
• Intelligent Workflow: Creates complete project workflows with critical paths and parallel execution tracks
• Smart Dependencies: Automatically manages task dependencies and prevents circular dependencies

Supported Roles:

• Product Manager, UI Designer, UX Designer
• Frontend Developer, Backend Developer, Fullstack Developer, Mobile Developer
• QA Engineer, DevOps Engineer, Tech Lead, Project Manager

When working with a new project, simply tell the Agent "init project rules". This will guide the Agent to generate a set of rules tailored to your project's specific requirements and structure.

To develop or update features, use the command "plan task [your description]". The system will reference the previously established rules, attempt to understand your project, search for relevant code sections, and propose a comprehensive plan based on the current state of your project.

During the planning process, Shrimp guides the Agent through multiple steps of thinking. You can review this process and provide feedback if you feel it's heading in the wrong direction. Simply interrupt and share your perspective - the Agent will incorporate your feedback and continue the planning process.

When you're satisfied with the plan, use "execute task [task name or ID]" to implement it. If you don't specify a task name or ID, the system will automatically identify and execute the highest priority task.

If you prefer to execute all tasks in sequence without manual intervention for each task, use "continuous mode" to automatically process the entire task queue.

Due to LLM token limits, context may be lost during lengthy conversations. If this occurs, simply open a new chat session and ask the Agent to continue execution. The system will pick up where it left off without requiring you to repeat the task details or context.

Shrimp Task Manager provides structured document organization with project-specific paths. All AI-generated documents (analysis, reflection, research, team collaboration) are automatically stored in organized directories:

$DATA_DIR/projects/$PROJECT_NAME/docs/
├── project/          # Core project docs (README, architecture, roadmap)
├── prd/              # Product requirements and milestones
├── tasks/            # Task-related documentation
│   ├── completed/    # Completed task summaries
│   ├── in-progress/  # In-progress task analysis
│   ├── pending/      # Pending task planning
│   └── by-role/      # Role-based classification (frontend, backend, qa, devops, design)
├── analysis/         # Analysis reports (complexity, risk, performance, security)
├── reflection/       # Reflection summaries (sprint reviews, lessons learned, improvements)
├── research/         # Research documents (tech research, competitor analysis, best practices)
├── team/             # Team collaboration docs (knowledge base, collaboration patterns, roles)
└── templates/        # Document templates

Key Benefits:

• Automatic Organization: Documents are automatically categorized by type and purpose
• Project Isolation: Each project maintains its own document structure
• Team Collaboration: Structured paths for team knowledge sharing and collaboration
• Easy Navigation: Clear hierarchy makes finding documents intuitive

Usage: The system automatically uses correct paths when creating documents. You can also explicitly request document paths using the get_documentation_path tool.

You can switch the language of system prompts by setting the TEMPLATES_USE environment variable. It supports en (English) and zh (Traditional Chinese) by default. Furthermore, you can copy an existing template directory (e.g., src/prompts/templates_en) to the location specified by DATA_DIR, modify it, and then point TEMPLATES_USE to your custom template directory name. This allows for deeper prompt customization. For detailed instructions.

Shrimp Task Manager includes a specialized research mode designed for systematic technical investigation and knowledge gathering.

Research Mode is a guided workflow system that helps AI Agents conduct thorough and systematic technical research. It provides structured approaches to exploring technologies, comparing solutions, investigating best practices, and gathering comprehensive information for programming tasks.

• Systematic Investigation: Structured workflows ensure comprehensive coverage of research topics
• Multi-Source Research: Combines web search and codebase analysis for complete understanding
• State Management: Maintains research context and progress across multiple sessions
• Guided Exploration: Prevents research from becoming unfocused or going off-topic
• Knowledge Integration: Seamlessly integrates research findings with task planning and execution

Research Mode is particularly valuable for:

• Technology Exploration: Investigating new frameworks, libraries, or tools
• Best Practices Research: Finding industry standards and recommended approaches
• Solution Comparison: Evaluating different technical approaches or architectures
• Problem Investigation: Deep-diving into complex technical challenges
• Architecture Planning: Researching design patterns and system architectures

Simply tell the Agent to enter research mode with your topic:

• Basic usage: "Enter research mode for [your topic]"
• Specific research: "Research [specific technology/problem]"
• Comparative analysis: "Research and compare [options A vs B]"

The system will guide the Agent through structured research phases, ensuring thorough investigation while maintaining focus on your specific needs.

1. Topic Definition: Clearly define the research scope and objectives
2. Information Gathering: Systematic collection of relevant information
3. Analysis and Synthesis: Processing and organizing findings
4. State Updates: Regular progress tracking and context preservation
5. Integration: Applying research results to your project context

💡 Recommendation: For the best research mode experience, we recommend using Claude 4 Sonnet, which provides exceptional analytical capabilities and comprehensive research synthesis.

Shrimp Task Manager has long-term memory capabilities, automatically saving task execution history and providing reference experiences when planning new tasks.

• The system automatically backs up tasks to the memory directory
• Backup files are named in chronological order, in the format tasks_backup_YYYY-MM-DDThh-mm-ss.json
• Task planning Agents automatically receive guidance on how to use the memory function

• Avoid Duplicate Work: Reference past tasks, no need to solve similar problems from scratch
• Learn from Successful Experiences: Utilize proven effective solutions, improve development efficiency
• Learning and Improvement: Identify past mistakes or inefficient solutions, continuously optimize workflows
• Knowledge Accumulation: Form a continuously expanding knowledge base as system usage increases

Through effective use of the task memory function, the system can continuously accumulate experience, with intelligence level and work efficiency continuously improving.

Building on the individual task memory system, Shrimp Task Manager includes advanced team collaboration memory features designed specifically for team environments.

• 🧠 Team Knowledge Sharing: Share experiences, solutions, and best practices across team members
• 📚 Collective Learning: Build a shared knowledge base that grows with team experience
• 🤝 Collaboration Pattern Learning: Automatically learn and recommend effective collaboration patterns
• 👥 Role-Based Knowledge Distribution: Tailor knowledge sharing based on team roles and expertise
• 📊 Team Performance Analytics: Analyze team collaboration effectiveness and identify improvement opportunities

Knowledge Sharing (share_team_knowledge):

• Share technical solutions and best practices
• Rate and validate team knowledge for quality assurance
• Record collaboration patterns and their effectiveness
• Document team learning experiences and lessons learned
• Query relevant knowledge based on role and context

Collaboration Analysis (analyze_team_collaboration):

• Analyze team performance and collaboration effectiveness
• Identify knowledge gaps and skill distribution
• Track learning trends and knowledge retention
• Evaluate role contributions and specializations
• Generate improvement recommendations and action plans

• Knowledge Continuity: New team members can quickly access accumulated team knowledge
• Avoid Repeated Mistakes: Learn from team's collective experience and past failures
• Optimize Collaboration: Data-driven insights to improve team workflows and communication
• Skill Development: Identify learning opportunities and knowledge gaps across the team
• Best Practice Propagation: Successful patterns and solutions spread naturally across the team

Sharing Knowledge:

"Share this solution with the team: [describe your solution]"
"Record this collaboration pattern as successful"
"Rate this team knowledge as helpful"

Analyzing Team Performance:

"Analyze our team's collaboration effectiveness"
"What are our knowledge gaps in frontend development?"
"Show me learning trends for the past quarter"

This team-oriented memory system transforms individual task execution into collective team intelligence, ensuring that knowledge and experience benefit the entire team rather than staying isolated with individual contributors.

The Project Rules feature helps maintain consistency across your codebase:

• Standardize Development: Establish consistent coding patterns and practices
• Onboard New Developers: Provide clear guidelines for project contributions
• Maintain Quality: Ensure all code meets established project standards

⚠️ Recommendation: Initialize project rules when your project grows larger or undergoes significant changes. This helps maintain consistency and quality as complexity increases.

Use the init_project_rules tool to set up or update project standards when:

• Starting a new large-scale project
• Onboarding new team members
• Implementing major architectural changes
• Adopting new development conventions

You can easily access this feature with simple natural language commands:

• For initial setup: Simply tell the Agent "init rules" or "init project rules"
• For updates: When your project evolves, tell the Agent "Update rules" or "Update project rules"

This tool is particularly valuable when your codebase expands or undergoes significant structural changes, helping maintain consistent development practices throughout the project lifecycle.

• Prompt Customization Guide: Instructions for customizing tool prompts via environment variables
• Changelog: Record of all notable changes to this project

To install Shrimp Task Manager for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @cjo4m06/mcp-shrimp-task-manager --client claude

# Install dependencies
npm install

# Build and start service
npm run build

Shrimp Task Manager includes a comprehensive testing suite powered by Vitest to ensure code quality and reliability.

# Run all tests
npm test

# Run tests in watch mode (interactive)
npm run test:watch

# Run tests with coverage report
npm run test:coverage

# Run tests with UI interface
npm run test:ui

# Run tests for CI (no watch, with coverage)
npm run test:ci

We use Vitest as our testing framework, which provides:

• Fast execution with native ES modules support
• Better TypeScript integration out of the box
• Modern testing features with excellent developer experience
• Built-in coverage reporting with v8 provider
• Watch mode with intelligent re-running

The project maintains a minimum test coverage threshold of 50% across:

• Lines, Functions, Branches, and Statements
• Core task management operations
• Tool functionality and utilities
• Type definitions and validation

Every pull request automatically runs:

• Comprehensive test suite on Node.js 18.x and 20.x
• TypeScript compilation validation
• Code coverage analysis with threshold enforcement
• Security audit for vulnerabilities
• Build verification to ensure deployability

Coverage reports are automatically posted as PR comments, and builds fail if coverage drops below the 50% threshold.

For contributors:

• Write tests for new features and bug fixes
• Ensure all tests pass before submitting PRs
• Maintain or improve code coverage
• Follow the established testing patterns
• Use Vitest's modern testing features and syntax

For detailed testing information, see TESTING.md.

Shrimp Task Manager can be used with any client that supports the Model Context Protocol, such as Cursor IDE.

Shrimp Task Manager offers two configuration methods: global configuration and project-specific configuration.

1. Open the Cursor IDE global configuration file (usually located at ~/.cursor/mcp.json)
2. Add the following configuration in the mcpServers section:

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "node",
      "args": ["/mcp-shrimp-task-manager/dist/index.js"],
      "env": {
        "DATA_DIR": "/path/to/project/data", // 必須使用絕對路徑
        "TEMPLATES_USE": "en",
        "ENABLE_GUI": "false"
      }
    }
  }
}


or

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "npx",
      "args": ["-y", "mcp-shrimp-task-manager"],
      "env": {
        "DATA_DIR": "/mcp-shrimp-task-manager/data",
        "TEMPLATES_USE": "en",
        "ENABLE_GUI": "false"
      }
    }
  }
}

⚠️ Please replace /mcp-shrimp-task-manager with your actual path.

You can also set up dedicated configurations for each project to use independent data directories for different projects:

1. Create a .cursor directory in the project root
2. Create an mcp.json file in this directory with the following content:

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "node",
      "args": ["/path/to/mcp-shrimp-task-manager/dist/index.js"],
      "env": {
        "DATA_DIR": "/path/to/project/data", // Must use absolute path
        "TEMPLATES_USE": "en",
        "ENABLE_GUI": "false"
      }
    }
  }
}


or

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "npx",
      "args": ["-y", "mcp-shrimp-task-manager"],
      "env": {
        "DATA_DIR": "/path/to/project/data", // Must use absolute path
        "TEMPLATES_USE": "en",
        "ENABLE_GUI": "false"
      }
    }
  }
}

The DATA_DIR parameter is the directory where Shrimp Task Manager stores task data, conversation logs, and other information. Setting this parameter correctly is crucial for the normal operation of the system. This parameter must use an absolute path; using a relative path may cause the system to incorrectly locate the data directory, resulting in data loss or function failure.

Warning: Using relative paths may cause the following issues:

• Data files not found, causing system initialization failure
• Task status loss or inability to save correctly
• Inconsistent application behavior across different environments
• System crashes or failure to start

Shrimp Task Manager supports project-aware data directories to automatically isolate task data for different projects. This feature prevents task conflicts when working with multiple projects simultaneously.

The system automatically detects your current project and creates project-specific data directories:

DATA_DIR/
├── projects/
│   ├── project-a/
│   │   ├── tasks.json
│   │   ├── memory/
│   │   └── WebGUI.md
│   ├── project-b/
│   │   ├── tasks.json
│   │   └── memory/
│   └── my-custom-project/
│       └── tasks.json
└── tasks.json (legacy, for backward compatibility)

1. PROJECT_NAME environment variable (highest priority)
2. Git repository name
3. package.json name field
4. Current working directory name (lowest priority)

Add these environment variables to enable project-aware functionality:

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "node",
      "args": ["/path/to/mcp-shrimp-task-manager/dist/index.js"],
      "env": {
        "DATA_DIR": "/path/to/global/data",
        "PROJECT_AUTO_DETECT": "true",
        "PROJECT_NAME": "my-custom-project",
        "TEMPLATES_USE": "en",
        "ENABLE_GUI": "false"
      }
    }
  }
}

Environment Variables:

• PROJECT_AUTO_DETECT: Set to "true" to enable project detection (default: "false")
• PROJECT_NAME: Optional manual project identifier (overrides automatic detection)

• Multi-Project Support: Work on multiple projects without task data conflicts
• Automatic Isolation: Each project gets its own task history and memory
• Backward Compatibility: Disabled by default, existing users unaffected
• Flexible Detection: Multiple detection methods ensure reliable project identification

Shrimp Task Manager supports customizing prompt behavior through environment variables, allowing you to fine-tune AI assistant responses without modifying code. You can set these variables in the configuration or through an .env file:

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "node",
      "args": ["/path/to/mcp-shrimp-task-manager/dist/index.js"],
      "env": {
        "DATA_DIR": "/path/to/project/data",
        "MCP_PROMPT_PLAN_TASK": "Custom planning guidance...",
        "MCP_PROMPT_EXECUTE_TASK_APPEND": "Additional execution instructions...",
        "TEMPLATES_USE": "en",
        "ENABLE_GUI": "false"
      }
    }
  }
}

There are two customization methods:

• Override Mode (MCP_PROMPT_[FUNCTION_NAME]): Completely replace the default prompt
• Append Mode (MCP_PROMPT_[FUNCTION_NAME]_APPEND): Add content to the existing prompt

Additionally, there are other system configuration variables:

• DATA_DIR: Specifies the directory where task data is stored
• PROJECT_AUTO_DETECT: Enable automatic project detection for data isolation (default: "false")
• PROJECT_NAME: Manual project identifier, overrides automatic detection
• TEMPLATES_USE: Specifies the template set to use for prompts. Defaults to en. Currently available options are en and zh. To use custom templates, copy the src/prompts/templates_en directory to the location specified by DATA_DIR, rename the copied directory (e.g., to my_templates), and set TEMPLATES_USE to the new directory name (e.g., my_templates).
• LOG_LEVEL: Controls the logging level. Available options: DEBUG, INFO, WARN, ERROR (default: "INFO"). Logs are automatically saved to DATA_DIR/logs/ directory.

For detailed instructions on customizing prompts, including supported parameters and examples, see the Prompt Customization Guide.

You can enable Cursor Settings => Features => Custom modes, and configure the following two modes:

You are a professional task planning expert. You must interact with users, analyze their needs, and collect project-related information. Finally, you must use "plan_task" to create tasks. When the task is created, you must summarize it and inform the user to use the "TaskExecutor" mode to execute the task.
You must focus on task planning. Do not use "execute_task" to execute tasks.
Serious warning: you are a task planning expert, you cannot modify the program code directly, you can only plan tasks, and you cannot modify the program code directly, you can only plan tasks.

You are a professional task execution expert. When a user specifies a task to execute, use "execute_task" to execute the task.
If no task is specified, use "list_tasks" to find unexecuted tasks and execute them.
When the execution is completed, a summary must be given to inform the user of the conclusion.
You can only perform one task at a time, and when a task is completed, you are prohibited from performing the next task unless the user explicitly tells you to.
If the user requests "continuous mode", all tasks will be executed in sequence.

💡 Choose the appropriate mode based on your needs:

• Use TaskPlanner mode when planning tasks
• Use TaskExecutor mode when executing tasks

If your tool doesn't support Custom modes, you can:

• Manually paste the appropriate prompts at different stages
• Or directly use simple commands like Please plan the following task: ...... or Please start executing the task...

After configuration, you can use the following tools:

• Node.js: High-performance JavaScript runtime environment
• TypeScript: Provides type-safe development environment
• MCP SDK: Interface for seamless interaction with large language models
• UUID: Generate unique and reliable task identifiers

This project is licensed under the MIT License - see the LICENSE file for details.

For the best experience, we recommend using the following models:

• Claude 3.7: Offers strong understanding and generation capabilities.
• Gemini 2.5: Google's latest model, performs excellently.

Due to differences in training methods and understanding capabilities across models, using other models might lead to varying results for the same prompts. This project has been optimized for Claude 3.7 and Gemini 2.5.