思维协议MCP服务器
](https://www.npmjs.com/package/thinking-protocol-mcp)  ](https://github.com/RimGit-N/thinking-protocol-mcp/stargazers) ](https://github.com/RimGit-N/thinking-protocol-mcp/issues) ](https://nodejs.org/) 
一种模型上下文协议(MCP)服务器,它实现了一个结构化的6阶段思维框架,用于系统地解决问题和决策。
🎯 概述
此MCP服务器提供了引导AI代理通过6个强制性阶段的严格思维过程的工具:
- 观察 -仅捕获目标输入
- 分析 -推导意义和上下文
- 根 -确定真正的根本原因
- 法案 -定义可操作的决策
- 验证 -验证结果和正确性
- 改进 -防止再次发生,加强制度
🚀 安装
选项1:通过NPM(推荐-发布后)
npm install -g thinking-protocol-mcp选项2:来自GitHub
git clone https://github.com/RimGit-N/thinking-protocol-mcp.git
cd thinking-protocol-mcp
npm install选项3:通过NPX(无需安装)
npx -y thinking-protocol-mcp🔧 配置
添加到您的MCP配置文件中(例如。, mcp_config.json 或Claude桌面配置):
使用NPX(推荐)
{
"mcpServers": {
"thinking-protocol": {
"command": "npx",
"args": [
"-y",
"thinking-protocol-mcp"
]
}
}
}使用全局安装
{
"mcpServers": {
"thinking-protocol": {
"command": "thinking-protocol-mcp",
"args": []
}
}
}使用本地克隆
{
"mcpServers": {
"thinking-protocol": {
"command": "node",
"args": [
"/absolute/path/to/thinking-protocol-mcp/index.js"
]
}
}
}多个MCP服务器的完整示例
如果您同时使用多个MCP服务器:
{
"mcpServers": {
"chrome-devtools": {
"command": "npx",
"args": [
"-y",
"chrome-devtools-mcp"
]
},
"context7": {
"command": "npx",
"args": [
"-y",
"@upstash/context7-mcp"
]
},
"thinking-protocol": {
"command": "npx",
"args": [
"-y",
"thinking-protocol-mcp"
]
}
}
}备注:确保有逗号(,)在每个服务器条目之间,除了最后一个条目之后。
🛠️ 可用工具
1. thinking_start_session
开始一个新的思考协议会议。
输入:
{
"task_description": "Description of the problem to solve"
}输出:
{
"success": true,
"session_id": "session_1234567890_abc123",
"next_stage": "observe",
"protocol": { ... }
}2. thinking_execute_stage
执行协议的特定阶段。
输入:
{
"session_id": "session_1234567890_abc123",
"stage": "observe",
"data": {
"facts": "Factual data...",
"signals": "Important indicators...",
"state": "Current system state..."
}
}阶段要求:
| 阶段 | 必填密钥 |
|---|---|
| 观察 | facts, signals, state |
| 分析 | patterns, assumptions, context |
| 根 | primary_cause, supporting_factors, core_issue |
| 行动 | main_action, alternatives_risks, execution_priority |
| 验证 | success_metrics, failure_signs, evaluation_method |
| 改善 | system_change, rules_sop, long_term_prevention |
3. thinking_get_status
获取思考会议的当前状态。
输入:
{
"session_id": "session_1234567890_abc123"
}4. thinking_get_protocol
获取完整的协议配置。
输入:
{}5. thinking_validate_session
验证所有阶段是否正确完成。
输入:
{
"session_id": "session_1234567890_abc123"
}📋 用法示例
// 1. Start session
const session = await thinking_start_session({
task_description: "Optimize database query performance"
});
// 2. Execute OBSERVE stage
await thinking_execute_stage({
session_id: session.session_id,
stage: "observe",
data: {
facts: "Query takes 5 seconds, 10M rows, no indexes",
signals: "CPU usage spikes to 100% during query",
state: "Production database, peak hours"
}
});
// 3. Execute ANALYZE stage
await thinking_execute_stage({
session_id: session.session_id,
stage: "analyze",
data: {
patterns: "Full table scans on large dataset",
assumptions: "No proper indexing strategy",
context: "Legacy system, no recent optimization"
}
});
// ... continue with remaining stages
// 6. Validate session
const validation = await thinking_validate_session({
session_id: session.session_id
});🔒 执行规则
- 模式:严格-必须完成所有阶段
- 已订购:TRUE-阶段必须按顺序执行
- 允许跳过:FALSE-不能跳过任何阶段
- 需要所有密钥:TRUE-必须填写所有必需的密钥
🎓 何时使用
强制要求:
- ✅ 复杂的问题解决
- ✅ 根本原因分析
- ✅ 架构决策
- ✅ 安全分析
- ✅ 性能优化
可选用于:
- ⚪ 简单、直接的任务
- ⚪ 常规操作
📄 许可证
麻省理工学院
👤 作者
里穆鲁
