jhawkins11/task-manager-mcp

开发工具

本页由 AI 翻译，技术细节请以原文为准。查看原文

21 stars 3 forks 质量 55 趋势 10

Repository Install

Node.js MCP服务器，使用LLM（Gemini/OpenRouter）进行AI辅助任务规划、拆解和代码审查。设计用于光标集成。

概览

这是一个MCP服务器，旨在与像Cursor这样的AI代码编辑器集成。这里的主要目标是最大化Cursor的代理能力和Gemini 2.5出色的架构能力，同时绕过Cursor极其有限的上下文窗口。这在很大程度上受到Roo Code的Boomerang模式启发，但我觉得它非常昂贵，因为唯一能支持其应用机器人的模型是Claude 3.7 Sonnet。有了这个服务器，你可以同时享受两全其美：无限上下文窗口和无限使用，而Cursor每月20美元的订阅费同样如此。此外，它还包含一个Svelte界面，允许你查看任务列表和进度，手动调整计划，并查看变更。 - 给它一个功能描述，它通过仓库使用带有项目上下文的LLM，生成逐步编码计划，供AI代理遵循，并对高强度任务进行递归拆解。

概述（AI 翻译）

此外，它还包含一个Svelte界面，允许你查看任务列表和进度，手动调整计划，并查看变更。

纤细的UI

核心功能

复杂特征规划： 给它一个特征描述，它通过repomix生成逐步编码计划，供AI代理遵循，并递归任务拆解高强度任务。
**集成UI服务器：**运行Express服务器以提供静态前端文件，并为用户界面提供基本API端点。

本页由 AI 翻译，技术细节请以原文为准。

README

以下为原文 README 节选，完整内容请查看原文。

Task Manager MCP Server

This is an MCP server built to integrate with AI code editors like Cursor. The main goal here is to maximize Cursor’s agentic capabilities and Gemini 2.5’s excellent architecting capabilities while working around Cursor’s extremely limited context window. This was inspired largely by Roo Code’s Boomerang mode, but I found it extremely expensive as the only model that works with it’s apply bot is Claude 3.7 Sonnet. With this server, you get the best of both worlds: unlimited context window and unlimited usage for the price of Cursor’s $20/month subscription.

In addition, it includes a Svelte UI that allows you to view the task list and progress, manually adjust the plan, and review the changes.

Svelte UI

Core Features

Complex Feature Planning: Give it a feature description, and it uses an LLM with project context via repomix to generate a step-by-step coding plan for the AI agent to follow with recursive task breakdown for high-effort tasks.
Integrated UI Server: Runs an Express server to serve static frontend files and provides basic API endpoints for the UI. Opens the UI in the default browser after planning is complete or when clarification is needed and displays the task list and progress.
Unlimited Context Window: Uses Gemini 2.5’s 1 million token context window with repomix’s truncation when needed.
Conversation History: Keeps track of the conversation history for each feature in a separate JSON file within .mcp/features/ for each feature, allowing Gemini 2.5 to have context when the user asks for adjustments to the plan.
Clarification Workflow: Handles cases where the LLM needs more info, pausing planning and interacting with a connected UI via WebSockets.
Task CRUD: Allows for creating, reading, updating, and deleting tasks via the UI.
Code Review: Analyzes git diff HEAD output using an LLM and creates new tasks if needed.
Automatic Review (Optional): If configured (AUTO_REVIEW_ON_COMPLETION=true), automatically runs the code review process after the last original task for a feature is completed.
Plan Adjustment: Allows for adjusting the plan after it’s created via the adjust_plan tool.

Setup

Prerequisites:

Node.js
npm
Git

Installation & Build:

Clone:

git clone https://github.com/jhawkins11/task-manager-mcp.git
cd task-manager-mcp

Install Backend Deps:
```
npm install
```
Configure: You’ll configure API keys later directly in Cursor’s MCP settings (see Usage section), but you might still want a local .env file for manual testing (see Configuration section).
Build: This command builds the backend and frontend servers and copies the Svelte UI to the dist/frontend-ui/ directory.
```
npm run build
```

Running the Server (Manually):

在 GitHub 查看完整 README

安装

This server does not publish a one-line install command.

Open the repository installation guide 新手？先看 mcpskill.org 教程 →

配置

{
  "mcpServers": {
    "task-manager-mcp": {
      "command": "node",
      "args": ["/full/path/to/your/task-manager-mcp/dist/server.js"],
      "env": {
        "OPENROUTER_API_KEY": "sk-or-v1-xxxxxxxxxxxxxxxxxxxx"
        //   optional: my recommended model for MCP is Gemini 2.5 Pro Free which is already set by default
        //   "OPENROUTER_MODEL": "google/gemini-2.5-flash-preview:thinking",
        //   also optional
        //   "FALLBACK_OPENROUTER_MODEL": "google/gemini-2.5-flash-preview:thinking",
        //   optional: the default port for the UI is 4999 if not set
        //   "UI_PORT": "4999",
        //   optional: the default port for the WebSocket server is 4999 if not set
        //   "WS_PORT": "4999"
        // Add GEMINI_API_KEY here instead if using Google directly
        // Add any other necessary env vars here
      }
    }
    // Add other MCP servers here if you have them
  }
}