gcrew-mcp

gcrew-mcp is a task orchestration and state management tool based on the Model Context Protocol (MCP). Its core philosophy is to serve as the "Orchestrator/Brain" for AI agents, responsible for decomposing complex goals into executable sub-tasks and improving work efficiency through state tracking.

Core Positioning

• gcrew-mcp (Orchestrator/Brain): Task dispatcher. Responsible for "planning" and "tracking". It helps AI agents manage complex task flows, ensuring every step is correctly recorded and obtains the required context.

Typical Use Case: Task Automation Orchestration

1. Task Definition: The user assigns a complex task.
2. Orchestrator Planning (gcrew-mcp):
   • Calls create_main_task: Define core goal and description.
   • Calls decompose_task: Decompose into multiple specific sub-tasks and specify dependencies.
3. Sub-task Execution:
   • Main Agent calls get_next_task to get the next executable task.
   • Main Agent calls get_launch_command to get the command for starting a sub-agent (containing results from prerequisite tasks).
   • Main Agent executes the command to start a new sub-agent instance for specific work.
   • Sub-agent completes the work, and the Main Agent calls update_task_status to update status and store results.

Key Features

1. Automatic Model Selection (Token Efficiency)

When decomposing tasks, gcrew-mcp automatically analyzes the complexity of each sub-task. For simple tasks (short descriptions without complex keywords like refactor or debug), it defaults to using gemini-1.5-flash to minimize token costs.

2. Flexible Task Focusing

The get_next_task tool now supports an optional taskId parameter. This allows users or agents to focus on a specific task thread, making it easier to manage multiple concurrent projects.

3. Smart Command Generation

get_launch_command generates ready-to-use CLI commands that automatically include:

• Specified model flags (-m).
• Complete task context.
• Results from all completed prerequisite tasks (Data Pipeline).

Installation & Execution

Development Mode

npm install
npm run dev

Build

npm run build

MCP Tools Description

• create_main_task: Create a high-level task.
• decompose_task: Decompose a task into sub-tasks. Automatically selects model if not specified.
• list_tasks: List all tasks and their status, including assigned models.
• update_task_status: Update the status of a main task or sub-task and store results.
• get_next_task: Get the next pending sub-task. Optionally filter by taskId.
• get_launch_command: Generate the launch command for a sub-agent (combining task context, model selection, and dependency results).

Development Goals & Roadmap

• [x] Basic Task Management (Create, Decompose, List, Update)
• [x] Sub-Agent Command Generation (Launch Command)
• [x] Execution Hint: Provide execution environment suggestions for sub-tasks.
• [x] Dependency Graph: Support dependsOn relationships between sub-tasks.
• [x] Data Pipeline: Sub-tasks store results, passed to subsequent tasks.
• [x] Auto Model Selection: Automatically assign cost-effective models for simple tasks.
• [ ] Reliability (Unit Tests): Establish automated tests for storage and index.
• [ ] Persistent Storage Optimization: Upgrade from JSON files to a more robust storage solution.