循环MCP服务器
](https://www.npmjs.com/package/cycles-mcp)  
一个模型上下文协议(MCP)服务器,用于管理所有存储库中的开发周期、任务和git工作流。
特性
- 🔄 基于周期的开发 -将工作组织成有时间限制的周期
- 📝 任务管理 -使用自动进度更新创建和跟踪任务
- 🎯 灵活的尺寸 -在简单层或粒度控制之间进行选择
- 🔧 Git集成 -直接提交、推送和创建PR
- 📊 进度跟踪 -自动进度条和会话日志
- 🔀 依赖追踪 -自动对并行执行的任务进行分组
- 👥 团队协调 -查看哪些任务可以同时完成
- 📄 模板 -所有项目的一致文档
- 🤖 AI友好 -与Cursor和其他MCP客户端无缝协作
安装
添加到MCP客户端设置(~/.cursor/mcp.json 对于光标, ~/.config/claude/claude_desktop_config.json 克劳德桌面版):
选项A:npx(推荐,零安装)
{
"mcpServers": {
"cycles": {
"command": "npx",
"args": ["cycles-mcp"]
}
}
}选项B:全局安装
npm install -g cycles-mcp{
"mcpServers": {
"cycles": {
"command": "cycles-mcp"
}
}
}方案C:地方发展
git clone https://github.com/btbishop93/cycles-mcp.git
cd cycles-mcp
bun install
bun run build{
"mcpServers": {
"cycles": {
"command": "node",
"args": ["/path/to/cycles-mcp/dist/index.js"]
}
}
}重新启动MCP客户端
重新启动Cursor/Claude Desktop以加载MCP服务器。
用法
初始化新回购中的工作流
初始化时,AI将提示您进行所有必需的配置:
User: Initialize the cycles workflow for this project
AI: I'll help you set up the workflow. First, let me ask a few questions:
- What sizing mode do you prefer? (simple/granular)
- [If simple] What skill tier? (junior/mid/senior)
- [If granular] What difficulty/duration/detail level?
- How long should each cycle be? (weeks/months/quarters)
- How many hours do you have per cycle?
AI: *calls init-workflow with your configuration*重要提示: AI将在创建工作流之前要求进行配置。您必须提供:
- 尺寸模式(简单或粒度)
- 技能等级/级别
- 周期持续时间和时间框架
- 每个周期可用小时数
简单模式 (推荐给大多数用户):
User: Initialize workflow with simple mode, mid tier, 2 weeks, 16 hours
AI: *calls init-workflow with your settings*粒度模式 (适用于高级用户):
User: Initialize workflow with 8h tasks, senior difficulty, low detail level, 1 month, 40 hours
AI: *calls init-workflow with granular settings*可用选项:
- 简单模式:选择一层
- junior -1小时任务,高细节(循序渐进) - mid -2小时任务,中等细节(平衡) - senior -4小时任务,低细节(高层次目标)
- 粒度模式:指定每个:
- 困难: junior, mid, senior - 持续时间: 0.5h, 1h, 2h, 4h, 8h - 详细信息: high, medium, low
- 循环周期:
- 1-3周 - 1-2个月 - 1个季度(3个月)
- 每个周期的小时数:你能为这个周期投入多少小时
创建循环
User: Create a new cycle called "Authentication System"
AI: *calls create-cycle*详细信息:
User: Create cycle "API Development" with goal "Build REST API"
AI: *calls create-cycle with description and goal*将任务添加到循环中
基本任务:
User: Add task "Setup database schema" to cycle 01
AI: *calls add-task with default settings from config*有依赖关系(用于并行化):
User: Add task "User authentication" to cycle 01, depends on Task 001
AI: *calls add-task with dependencies, automatically groups tasks*使用完整元数据:
User: Add task "Implement OAuth" to cycle 01:
- Depends on: Task 001, Task 002
- Conflicts with: None
- Modifies: src/auth/, database users table
- Duration: 8h, senior difficulty
AI: *calls add-task with full metadata*为什么要指定依赖关系?
- 任务会自动分组到并行执行组中
- 团队可以看到哪些任务可以同时完成
- LLM代理可以在不同的组中工作而不会发生冲突
- 循环README显示了清晰的视觉分组:🟢 第1组(现在开始),🟡 第2组(在第1组之后)等。
处理任务
1.承诺你的工作
User: Commit this with message "add user model"
AI: *calls commit-task*详细正文:
User: Commit "add user model" with details about validation logic
AI: *calls commit-task with body*2.推送你的分支
User: Push this branch
AI: *calls push-branch*3.创建拉取请求
User: Create PR for task 001 "Setup database"
AI: *calls create-pr, automatically marks task complete*跟踪进度
User: Update progress for cycle 01, mark task 002 complete
AI: *calls update-progress*记录会话:
User: Log session for today, worked 2 hours on task 003
AI: *calls update-progress with session details*可用工具
| 工具 | 说明 |
|---|---|
init-workflow | 在存储库中初始化循环工作流 |
create-cycle | 创建新的开发周期 |
add-task | 将任务添加到循环中 |
commit-task | 使用常规提交格式提交更改 |
push-branch | 将当前分支推送到远程 |
create-pr | 创建拉取请求并标记任务完成 |
update-progress | 更新周期进度和记录会话 |
可用资源
| 资源 | 描述 |
|---|---|
template://workflow | WORKFLOW.md模板 |
template://cycles | cycles.md模板 |
template://cycle-readme | 循环README模板 |
template://task | 任务文件模板 |
template://pr | 拉取请求模板 |
工作流概述
1.初始化(每次重复一次)
重要: 这必须先做!所有其他工具都验证初始化是否完成。
init-workflow → creates:
- .cycles-config.json (configuration)
- WORKFLOW.md (workflow guide)
- docs/cycles.md (cycles tracker)
- docs/cycles/ (cycles directory)
- .github/pull_request_template.md (PR template)如果没有正确的初始化,其他工具将返回一个错误,列出丢失的文件。
2.创建循环
create-cycle → creates docs/cycles/01-cycle-name/README.md3.添加任务
add-task → creates docs/cycles/01-cycle-name/001-task-name.md
add-task → creates docs/cycles/01-cycle-name/002-task-name.md
...
→ Automatically updates cycle README with task dependencies
→ Groups tasks for parallel execution任务依赖性分组:
- 无依赖关系的任务→ 🟢 第1组(立即开始)
- 任务取决于第1组→ 🟡 第2组(平行于第1组)
- 任务取决于第2组→ 🟡 第3组(平行于第2组)
- 最终集成任务→ 🔴 最后一组
同一组内的任务可以同时处理!
4.按任务工作
# For each task:
1. Create branch: feat/cycle-01-task-001-description
2. Work and commit incrementally (commit-task)
3. Push branch (push-branch)
4. Create PR (create-pr) → auto-marks task complete
5. Repeat for next task5.跟踪进度
创建PR时,进度会自动更新。您还可以手动更新:
update-progress → marks tasks complete, updates progress bars配置
每个存储库都有一个 .cycles-config.json 文件:
{
"sizing_mode": "simple",
"simple_tier": "mid",
"cycle_duration": {
"unit": "weeks",
"value": 2
},
"hours_per_cycle": 16
}或颗粒状:
{
"sizing_mode": "granular",
"difficulty": "senior",
"task_duration": "4h",
"detail_level": "low",
"cycle_duration": {
"unit": "months",
"value": 1
},
"hours_per_cycle": 40
}例子
示例1:快速启动(简单模式)
User: Initialize cycles workflow with mid tier
AI: Creates structure, uses 2h tasks with medium detail
User: Create cycle "User Management"
AI: Creates cycle 01
User: Add tasks for signup, login, and profile
AI: Creates 3 tasks, each ~2h
User: (work on task 001, then...)
User: Commit "implement signup form"
AI: Commits with conventional format
User: Create PR for task 001
AI: Creates PR, marks task complete示例2:高级开发人员(粒度模式)
User: Initialize workflow with 8h tasks, senior difficulty, low detail
AI: Creates structure for experienced devs
User: Create cycle "Microservices Architecture"
AI: Creates cycle 01
User: Add task "Design service boundaries"
AI: Creates high-level task with minimal hand-holding
User: (work on task, then...)
User: Create PR for task 001
AI: Creates PR with your changes示例3:兼职贡献者
User: Initialize workflow with junior tier, 1 week cycles, 8 hours per cycle
AI: Creates structure with 8 1-hour tasks per cycle
User: Create cycle "Bug Fixes"
AI: Creates cycle with appropriate size
User: Add 8 small bug fix tasks
AI: Creates detailed, 1-hour tasks示例4:并行工作的团队
User: Initialize workflow for team collaboration
AI: Sets up structure
User: Create cycle "E-commerce Platform"
AI: Creates cycle 01
User: Add tasks:
1. "Setup database" - no dependencies
2. "Create product API" - depends on Task 001
3. "Design UI components" - no dependencies
4. "Build shopping cart" - depends on Task 002, 003
5. "Payment integration" - depends on Task 004
AI: Creates tasks and groups them automatically in cycle README:
🟢 Group 1 (Start Immediately):
- [ ] 001 - Setup database
- [ ] 003 - Design UI components
🟡 Group 2 (After Group 1):
- [ ] 002 - Create product API (needs 001)
🟡 Group 3 (After Group 2):
- [ ] 004 - Build shopping cart (needs 002, 003)
🔴 Group 4 (Final Tasks):
- [ ] 005 - Payment integration (needs 004)
> Parallelization tip: Tasks within the same group can be worked on simultaneously结果:
- 两个开发人员可以同时处理任务001和003
- 一旦完成,一个开发人员启动002,而另一个则处理文档
- 任务004在002和003完成后可用
- 整个团队清晰可见!
Git工作流集成
MCP服务器遵循以下约定:
分支命名
feat/cycle-XX-task-YYY-description例子: feat/cycle-01-task-001-setup-database
提交消息
(cycle-XX):
[optional body]例子: feat(cycle-01): add user authentication
拉取请求
- 标题:
feat(cycle-XX): Task description - 正文:根据PR模板自动生成
- 创建时自动将任务标记为已完成
发展
在开发模式下运行
bun run dev类型检查
bun run type-check构建
bun run build故障排除
服务器未显示在光标中
- 检查MCP设置文件是否为有效的JSON
- 确保路径
dist/index.js是正确的 - 完全重新启动游标
- 检查游标日志是否有错误
工具不工作
- 确保
init-workflow先跑 - 所有工具在运行之前都会验证工作流是否已正确初始化
- 所需文件:
.cycles-config.json,WORKFLOW.md,docs/cycles.md,docs/cycles/,.github/pull_request_template.md - 如果您看到“缺少文件”错误,请运行
init-workflow建立完整的结构 - 验证您是否在正确的目录中
- 检查文件权限
公关创建失败
MCP服务器尝试使用GitHub CLI(gh)创建PR。如果未安装:
- 安装GitHub命令行界面:
brew install gh - 身份验证:
gh auth login - 或者使用提供的模板手动创建PR
哲学
此MCP服务器基于 爱它或恨它 哲学:
- 无需定制 -模板是固定的,以保持一致性
- 意见化工作流程 -一种做事的方法,做得好
- 自动跟踪 -进度自动更新
- AI优先 -专为自然语言交互而设计
如果你需要定制,分叉和修改。否则,拥抱这种结构,享受生产力的提升!
许可证
麻省理工学院
贡献
问题和PR欢迎!这是一个爱好项目,但保持积极。
学分
创建用于管理跨多个存储库的开发周期。灵感来自敏捷冲刺,但针对单人开发人员和小团队进行了优化。
