MCP Task Splitting Server

A Model Context Protocol (MCP) server that provides intelligent task decomposition functionality. It guides models through a multi-step process to break down complex tasks into actionable subtasks.

Features

• Multi-step Task Refinement: Guides through 4 iterations to progressively refine tasks
• English Prompts: All system prompts are in English for better compatibility
• Smart Task Validation: Ensures subtasks fully achieve the main goal
• Progress Tracking: Tracks completion status of each subtask
• Flexible Task Types: Adapts decomposition based on task type (API, Frontend, Database, etc.)

Get Start

install

pnpm i task-splitting-mcp -g

Streamable HTTP Server

1. start http server:

task-splitting-mcp http

2. config file:

{
  "servers": {
    "task-splitting-mcp": {
      "url": "http://localhost:3001/mcp",
      "type": "http"
    }
  },
  "inputs": []
}

Stdio

{
  "servers": {
    "task-splitting-mcp": {
      "command": "task-splitting-mcp",
      "args": ["stdio"]
    }
  }
}

Running

If you see the following output, it means you have successfully started and configured the task-splitting-mcp mcp:

Usage

Task Splitting Feature

The server provides three main tools for task decomposition:

1. split_task - Breaks down a main task into subtasks through iterative refinement
2. get_task - Retrieves a specific subtask by index
3. check_task - Marks a subtask as complete and guides to the next one

Example Usage:

Please split the task "Build a REST API with user authentication" into subtasks

The server will guide you through 4 iterations:

• Iteration 0: Initial task breakdown
• Iteration 1: Refine tasks for specificity
• Iteration 2: Validate tasks against the main goal
• Iteration 3: Store final tasks and complete

See examples/task-splitting-demo.md for detailed examples.

Using the MCP Server Prompt

The server provides a built-in MCP prompt that can be called to automatically guide the model to use the task splitting functionality.

Built-in Prompt: task-splitting-mcp

This prompt accepts a task description and returns a message that guides the model to use the split_task tool:

Usage:

{
  "method": "prompts/get",
  "params": {
    "name": "task-splitting-mcp",
    "arguments": {
      "taskDescription": "Build a REST API with user authentication"
    }
  }
}

Response: The prompt returns a message asking the model to use the split_task tool with the provided task description.

Manual Integration

For manual integration, you can use this system prompt:

When users ask me to complete complex tasks, I will use the split_task tool to break them down into 4-6 subtasks through iterative refinement, then systematically complete each subtask using get_task and check_task.