sequential-thinking-mcp-v2

概述

Sequential Thinking MCP v2 是一个本地 MCP 服务器，实现了结构化推理、会话持久化和项目管理功能。它遵循第一性原理思维框架，并采用分层内存架构的已批准设计。服务器通过临时推理、检查点摘要、提升结论和规范文档来管理推理生命周期。计划在无法访问网络时采用云辅助操作和本地优先的回退方案。

试试问 AI

装完之后，这里有 7 个你可以让 AI 做的事：

• retrieve_context

  Retrieve relevant context for the current task

  试试:Retrieve relevant context for this coding task试试:Find context related to authentication
• search_docs

  Search through canonical documentation

  试试:Search for API documentation试试:Find design documents related to authentication
• search_memory

  Search through promoted memory artifacts

  试试:Search previous conclusions about performance issues试试:Find memory related to database optimization
• promote_memory

  Promote reasoning to durable memory

  试试:Promote this conclusion to memory for future reuse试试:Save this investigation summary as a memory artifact
• summarize_session

  Create a summary of the current session

  试试:Summarize this session's progress试试:Create a checkpoint summary of the work done
• generate_handoff

  Generate a handoff bundle for session continuity

  试试:Create a handoff document for this session试试:Generate a summary for another agent to continue this work
• map_changed_modules

  Map modules that have changed during the session

  试试:Show me which modules have been modified试试:Map changes to the codebase since last checkpoint

可对比工具

安装

git clone https://github.com/spideynolove/sequential-thinking-mcp-v2.git
cd sequential-thinking-mcp-v2

uv sync
uv run main.py

{
  "mcpServers": {
    "sequential-thinking": {
      "command": "python",
      "args": ["main.py"],
      "cwd": "/path/to/sequential-thinking-mcp-v2"
    }
  }
}

FAQ

内存架构设计是什么？

设计使用 Markdown 作为规范的人类真理，qmd 用于文档检索，Zvec 用于提升的工件，结构化元数据存储用于生命周期状态。

令牌管理如何工作？

项目以 64K 的活跃上下文上限为目标，在 45K 处设置压缩触发器。它按需检索，在子任务边界检查点，并压缩旧上下文。

网络访问失败时会发生什么？

服务器采用云辅助、本地优先的回退方法，确保即使在无法访问互联网时也能正常运行。

sequential-thinking-mcp-v2 对比