home / mcp / task manager mcp server

Task Manager MCP Server

Installation

Add the following to your MCP client configuration file.

Configuration

{
  "mcpServers": {
    "blizzy78-mcp-task-manager": {
      "command": "npx",
      "args": [
        "-y",
        "@blizzy/mcp-task-manager"
      ],
      "env": {
        "SINGLE_AGENT": "true"
      }
    }
  }
}

This Task Manager MCP Server lets you orchestrate multi-step task workflows, track progress, and manage complex task hierarchies across agents. It provides structured task creation, decomposition, and dependency-aware execution to help you coordinate work efficiently.

How to use

You will interact with the Task Manager MCP Server through a client that supports MCP tooling. Start by creating top-level tasks, then decompose large tasks into subtasks with explicit sequence ordering. Update tasks as you begin and complete work, and rely on the defined task states to track progress from todo to in-progress and finally to done or failed. Use the dependency and uncertainty information to guide parallel work where possible and ensure all critical-path tasks are completed before considering a project finished.

Key operations you will perform include creating tasks, decomposing complex tasks, updating statuses, requesting task details, and querying currently in-progress work when operating in single-agent mode. You should always include a verification step in each subtask and propagate dependency chains so later subtasks depend on earlier ones.

How to install

Prerequisites: you need a runtime capable of running Node.js commands (npm or npx available). The server runs as a local stdio process and can be started directly or via an editor’s MCP integration.

Install and run the server using the following command to start via npx (default stdio server):

npx @blizzy/mcp-task-manager

Single Agent Mode

If you want a single agent running in a dedicated context, enable single-agent mode. This adds the current_task endpoint for recovering in-progress work after context compaction.

Enable single-agent mode from the command line by setting SINGLE_AGENT=true before starting the server:

SINGLE_AGENT=true npx @blizzy/mcp-task-manager

Configuration for Claude Desktop or VS Code clients

If you are configuring Claude Desktop, specify the MCP server with standard stdio transport options and include the environment flag to enable single-agent mode when desired. Here is an example configuration block for Claude Desktop or a VS Code MCP setup.

Notes and best practices

- Disable the Todo List tool in VS Code when using this MCP server to avoid conflicting task views.

- All top-level tasks start in the todo state and must be updated before execution begins. Decompose complex tasks before execution when required.

Additional sections

The server supports structured task management tools that enable creation, decomposition, updating, and querying of tasks. Use these tools to build detailed task trees, define definitions of done, and capture uncertainty areas to guide your work.

Available tools

create_task

Creates a new task with a title, description, goal, and definitions of done. Starts in todo and must be updated before execution. If the task is complex, decompose it first.

decompose_task

Breaks a complex task into smaller subtasks with sequence ordering. Subtasks in the same sequence order can run in parallel. Each subtask includes a verification step and dependencies are created based on sequence.

update_task

Updates the status or properties of one or more tasks. Enforces valid transitions, records lessons learned, and requires that all critical path dependencies are completed before marking a task as done.

task_info

Returns full details for requested tasks, including status, dependencies, descriptions, and other properties. Handles missing IDs gracefully.

current_task

In single-agent mode, returns tasks currently in progress to help the agent recover context after history compaction.