使用说明

MCP顺序思维服务器

MCP服务器实现，通过结构化思维过程提供动态和反思性问题解决的工具。

特性

将复杂问题分解为可管理的步骤
随着理解的加深，修改和完善想法
分支到其他推理路径
动态调整想法总数
生成并验证解决方案假设

技术栈

FastMCP -MCP服务器框架
python -核心语言
紫外线 -包管理

可用工具

工具	说明
`create_thought_chain`	启动新的推理链
`add_thought`	为现有链添加一个想法
`revise_thought`	更新现有想法
`delete_thought`	从链中删除一个想法
`get_chain`	检索思维链
`summarize_chain`	获取推理摘要
`branch_chain`	创建备选推理路径
`add_hypothesis`	提出解决方案假设
`verify_hypothesis`	标记假设已验证/被拒绝
`list_chains`	列出所有活动链

安装

# Clone the repository
git clone https://github.com/ShawnKyzer/mcp-sequential-thinking.git
cd mcp-sequential-thinking

# Install dependencies with uv
uv sync

# Install with dev dependencies for testing
uv sync --all-extras

用法

运行服务器

# Run with stdio transport (default, for local MCP clients)
uv run mcp-sequential-thinking

# Or using the FastMCP CLI
uv run fastmcp run src/mcp_sequential_thinking/server.py:mcp

# Run with HTTP transport for remote access
uv run fastmcp run src/mcp_sequential_thinking/server.py:mcp --transport http --port 8000

Claude桌面配置

添加到您的Claude桌面配置(~/.config/claude/claude_desktop_config.json):

{
  "mcpServers": {
    "sequential-thinking": {
      "command": "uv",
      "args": ["run", "--directory", "/path/to/mcp-sequential-thinking", "mcp-sequential-thinking"]
    }
  }
}

程序化使用

from mcp_sequential_thinking import create_server, ThoughtManager, InMemoryThoughtChainStore

# Create a server with custom configuration
store = InMemoryThoughtChainStore()
server = create_server(name="My Thinking Server", store=store)

# Run the server
server.run()

发展

运行测试

# Run all tests
uv run pytest

# Run with verbose output
uv run pytest -v

# Run with coverage
uv run pytest --cov=mcp_sequential_thinking

# Run specific test file
uv run pytest tests/test_thought_manager.py -v

代码质量

该项目遵循Python最佳实践，包括：

坚实的原则：单一职责、开放/封闭、Liskov替换、接口隔离、依赖倒置
类型提示：贯穿全文的完整类型注释
Pydantic模型：经过验证的数据结构
异步/等待：非阻塞I/O操作
依赖注入：灵活的组件组成

项目结构

mcp-sequential-thinking/
├── pyproject.toml
├── README.md
├── src/
│   └── mcp_sequential_thinking/
│       ├── __init__.py
│       ├── server.py
│       ├── thought_manager.py
│       ├── models.py
│       └── store/
│           ├── __init__.py
│           ├── base.py
│           └── memory.py
├── tests/
└── design/
    └── ARCHITECTURE.md

用例

1.复杂代码架构决策

在设计新的系统架构时，人工智能可以使用顺序思维来：

Chain: "Microservices vs Monolith Decision"
├── Step 1: Analyze current team size and expertise
├── Step 2: Evaluate scalability requirements
├── Step 3: Consider deployment complexity
├── Step 4: Assess cost implications
├── Hypothesis: "Microservices better for this use case"
├── Step 5: Identify potential challenges
└── Verify: Hypothesis confirmed with evidence

示例提示： “帮助我为我们的电子商务平台在微服务和单体之间做出选择”

2.调试复杂问题

将调试分解为系统步骤：

Chain: "Memory Leak Investigation"
├── Step 1: Identify symptoms (high memory usage over time)
├── Step 2: Review recent code changes
├── Step 3: Analyze heap dumps
│   └── Branch: "Investigate connection pool"
├── Step 4: Narrow down to specific module
├── Hypothesis: "Unclosed database connections"
├── Step 5: Write test to reproduce
└── Verify: Hypothesis confirmed - fix applied

示例提示： “我们的Node.js应用程序在生产环境中泄漏内存，请帮我调查”

3.算法设计与优化

有条不紊地进行算法开发：

Chain: "Search Algorithm Optimization"
├── Step 1: Define problem constraints (1M records, <100ms)
├── Step 2: Evaluate brute force approach - O(n)
├── Step 3: Consider binary search - O(log n)
│   └── Branch: "What if data is unsorted?"
├── Step 4: Explore hash-based approach - O(1)
├── Hypothesis: "Hash map with caching optimal"
├── Step 5: Analyze memory trade-offs
├── Revise Step 4: Add LRU eviction for memory bounds
└── Verify: Benchmarks confirm 50ms average

示例提示： “使用自动补全功能为我们的产品目录设计高效搜索”

4.安全漏洞评估

系统安全分析：

Chain: "API Security Audit"
├── Step 1: Map all API endpoints
├── Step 2: Check authentication mechanisms
├── Step 3: Analyze authorization logic
├── Hypothesis: "IDOR vulnerability in /users/{id}"
├── Step 4: Test hypothesis with different user contexts
├── Step 5: Review rate limiting
│   └── Branch: "Check for enumeration attacks"
├── Verify: IDOR confirmed - recommend fix
└── Step 6: Document findings and remediation

示例提示： “针对安全漏洞审核我们的REST API”

5.数据库模式设计

迭代模式细化：

Chain: "E-commerce Database Schema"
├── Step 1: Identify core entities (User, Product, Order)
├── Step 2: Define relationships
├── Step 3: Normalize to 3NF
├── Hypothesis: "Denormalize order_items for read performance"
├── Step 4: Consider query patterns
│   └── Branch: "Evaluate NoSQL for product catalog"
├── Revise Step 3: Add strategic denormalization
├── Step 5: Plan indexing strategy
└── Verify: Query benchmarks meet SLA

示例提示： “为多供应商市场设计数据库模式”

6.代码审查和重构

结构化代码改进：

Chain: "Legacy Code Refactoring"
├── Step 1: Understand current functionality
├── Step 2: Identify code smells (God class, 2000 lines)
├── Step 3: Map dependencies
├── Hypothesis: "Extract 3 service classes"
├── Step 4: Plan incremental refactoring steps
│   └── Branch: "Consider strangler fig pattern"
├── Step 5: Write tests for current behavior
├── Step 6: Execute refactoring with test coverage
└── Verify: All tests pass, complexity reduced 60%

示例提示： “帮助我重构这个2000行UserService类”

7.技术决策文件

创建可审计的决策记录：

Chain: "Frontend Framework Selection"
├── Step 1: Define requirements (SSR, TypeScript, team skills)
├── Step 2: Evaluate React/Next.js
├── Step 3: Evaluate Vue/Nuxt
├── Step 4: Evaluate Svelte/SvelteKit
│   └── Branch: "Consider learning curve impact"
├── Hypothesis: "Next.js best fits requirements"
├── Step 5: Prototype critical features in Next.js
├── Step 6: Gather team feedback
└── Verify: Team consensus achieved, Next.js selected

示例提示： “帮助我们为仪表板选择前端框架”

8.事件响应和RCA

根本原因分析工作流程：

Chain: "Production Outage RCA"
├── Step 1: Document timeline of events
├── Step 2: Identify affected systems
├── Step 3: Analyze logs and metrics
├── Hypothesis: "Database connection pool exhausted"
├── Step 4: Correlate with recent deployments
│   └── Branch: "Check for query performance regression"
├── Step 5: Identify contributing factors
├── Verify: Connection leak in new feature confirmed
└── Step 6: Document preventive measures

示例提示： “帮我对昨晚的停电进行根因分析”

运作原理

人工智能创造思维链 在处理复杂问题时
每个推理步骤都被记录下来 带有时间戳和元数据
思想可以修正 随着理解的加深
分支机构探索替代方案 而不会失去主线
提出并验证了假设 有证据
完整的审计跟踪 可供复习和学习

许可证

麻省理工学院