MCP Outcomes

@aegis ai/mcp结果

用于跟踪决策及其结果的MCP（模型上下文协议）服务器。记录你尝试了什么，它是否有效，并随着时间的推移获得数据驱动的建议。

安装

克劳德桌面版

添加到您的 claude_desktop_config.json:

{
  "mcpServers": {
    "outcomes": {
      "command": "bunx",
      "args": ["@aegis-ai/mcp-outcomes"]
    }
  }
}

或者使用npx：

{
  "mcpServers": {
    "outcomes": {
      "command": "npx",
      "args": ["-y", "@aegis-ai/mcp-outcomes"]
    }
  }
}

手册

git clone https://github.com/aegis-ai-coop/mcp-outcomes.git
cd mcp-outcomes
bun install
bun src/index.ts

工具

记录结果

记录决策及其结果。

{
  "decision": "Used Redis caching for session storage",
  "context": "infrastructure",
  "outcome": "Reduced p99 latency from 200ms to 15ms",
  "success": true,
  "tags": ["caching", "redis", "performance"]
}

查询结果

按标签、日期范围、文本或成功状态搜索过去的结果。

{
  "tags": ["performance"],
  "success": true,
  "limit": 20
}

{
  "text": "caching",
  "start_date": "2025-01-01",
  "end_date": "2025-06-30"
}

get_stats

返回成功率、每个标签的细分、趋势分析和建议。无需参数。

{}

示例响应：

{
  "total_outcomes": 47,
  "overall_success_rate": 0.72,
  "by_tag": {
    "caching": { "total": 12, "successes": 10, "failures": 2, "success_rate": 0.83 },
    "refactoring": { "total": 8, "successes": 3, "failures": 5, "success_rate": 0.38 }
  },
  "recent_trend": {
    "last_10_success_rate": 0.80,
    "last_30_success_rate": 0.70
  },
  "recommendations": [
    "Tag \"refactoring\" has a 38% success rate across 8 outcomes. Consider changing your approach.",
    "Tag \"caching\" performs well at 83% success across 12 outcomes. Keep using this approach.",
    "Recent performance is improving: last 10 at 80% vs last 30 at 70%. Current approach is working."
  ]
}

list_recent

列出N个最新结果。

{
  "count": 5
}

数据存储

结果存储在 ~/.mcp-outcomes/data.json。该文件在首次使用时自动创建。

特性

• 使用上下文、结果、标签和可选元数据记录决策
• 跨决策、上下文和结果的全文搜索
• 按标签、日期范围和成功/失败进行筛选
• 每个标签的成功率细分
• 比较近期与历史表现的趋势分析
• 基于数据模式的自动推荐
• 无外部依赖的本地JSON存储

许可证

麻省理工学院-AEGIS人工智能合作