Shrimp Task Manager

MCP Shrimp Task Manager

🚀 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.

✨ Features

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
Thought Chain Process: Step-by-step reasoning to analyze complex problems systematically
Project Rules Initialization: Define project standards and rules to maintain consistency across large projects

🧭 Usage Guide

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

What is Shrimp?

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.

First-Time Setup

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.

Task Planning Process

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.

Feedback Mechanism

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.

Task Execution

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.

Continuous Mode

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.

Token Limitation Note

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.

Prompt Language and Customization

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.

🧠 Task Memory Function

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

Key Features

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

Advantages and Benefits

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.

🤔 Thought Chain Process

The Thought Chain feature enhances problem-solving through structured thinking:

Systematic Reasoning: Break down complex problems into logical steps
Assumption Testing: Challenge assumptions to validate solution approaches
Critical Analysis: Evaluate solution options with rigorous criteria
Improved Decision Making: Reach more reliable conclusions through deliberate thinking

When enabled (default setting), the system guides the Agent through step-by-step reasoning using the process_thought tool, ensuring thorough problem analysis before implementation.

📋 Project Rules Initialization

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

Usage Examples

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.

📚 Documentation Resources

System Architecture: Detailed system design and data flow explanation
Prompt Customization Guide: Instructions for customizing tool prompts via environment variables
Changelog: Record of all notable changes to this project

🔧 Installation and Usage

Installing via Smithery

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

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

Manual Installation

# Install dependencies
npm install

# Build and start service
npm run build

🔌 Using with MCP-Compatible Clients

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

Configuration in Cursor IDE

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

Global Configuration

Open the Cursor IDE global configuration file (usually located at ~/.cursor/mcp.json)
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
      }
    }
  }
}


or

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

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

Project-Specific Configuration

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

Create a .cursor directory in the project root
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
      }
    }
  }
}


or

{
  "mcpServers": {
    "shrimp-task-manager": {
      "command": "npx",
      "args": ["-y", "mcp-shrimp-task-manager"],
      "env": {
        "DATA_DIR": "/path/to/project/data" // 必須使用絕對路徑
      }
    }
  }
}

⚠️ Important Configuration Notes

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

🔧 Environment Variable Configuration

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...",
        "ENABLE_THOUGHT_CHAIN": "true",
        "TEMPLATES_USE": "en"
      }
    }
  }
}

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
ENABLE_THOUGHT_CHAIN: Controls the thinking model in task planning workflow. When set to true (default), the system guides users to use the process_thought tool for step-by-step reasoning. When set to false, the system directly uses analyze_task to submit analysis results, skipping the detailed thinking process.
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).

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

💡 System Prompt Guidance

Cursor IDE Configuration

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

TaskPlanner Mode

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.

TaskExecutor Mode

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

Using with Other Tools

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...

🛠️ Available Tools Overview

After configuration, you can use the following tools:

Category	Tool Name	Description
Task Planning	`plan_task`	Start planning tasks
Task Analysis	`analyze_task`	In-depth analysis of task requirements
	`process_thought`	Step-by-step reasoning for complex problems
Solution Assessment	`reflect_task`	Reflect and improve solution concepts
Project Management	`init_project_rules`	Initialize or update project standards and rules
Task Management	`split_tasks`	Break tasks into subtasks
	`list_tasks`	Display all tasks and status
	`query_task`	Search and list tasks
	`get_task_detail`	Display complete task details
	`delete_task`	Delete incomplete tasks
Task Execution	`execute_task`	Execute specific tasks
	`verify_task`	Verify task completion
	`complete_task`	Mark tasks as completed

🔧 Technical Implementation

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

📄 License

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

Recommended Models

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.

Integrations

目錄