agent-loop-detector
Cloudflare Workers MCP server: agent-loop-detector
我该使用它吗
质量与安全性
A
基于对工具定义和协议合规性的自动分析。
上下文开销
~256token 数（工具定义）
~612 B典型响应大小
对注意力的影响极小（占 128k 上下文窗口的 0.20%）
这是每次将服务器的工具加载到模型上下文窗口时所消耗的大致 token 数。数值越高，可用于其他任务的注意力就越少。
安装
一键安装
将以下内容添加到你的 `claude_desktop_config.json` 文件中：
{
"mcpServers": {
"agent-loop-detector": {
"url": "https://api.lazy-mac.com/agent-loop-detector/mcp"
}
}
}远程端点
https://api.lazy-mac.com/agent-loop-detector/mcpstreamable-http它能做什么
工具清单
工具（3）
🟢 只读🟡 写入🔴 删除⚪ 未知
🟢analyze_trace(trace, pricing)
Full diagnostic: loop detection, cost breakdown, stuck-state classification, and fix recommendations for an AI agent trace.
输入模式
{
"type": "object",
"properties": {
"trace": {
"type": "array",
"description": "Ordered array of agent steps (role, tool, input, output, tokens, model)",
"items": {
"type": "object"
}
},
"pricing": {
"type": "object",
"nullable": true,
"properties": {
"in": {
"type": "number"
},
"out": {
"type": "number"
}
}
}
},
"required": [
"trace"
]
}⚪detect_loops(trace)
Detect exact-call repeats, periodic tool-call cycles, and ping-pong patterns in an agent trace.
输入模式
{
"type": "object",
"properties": {
"trace": {
"type": "array",
"items": {
"type": "object"
}
}
},
"required": [
"trace"
]
}🟢breakdown_cost(trace, pricing)
Calculate per-model token usage and USD cost from an agent trace. Supports custom pricing overrides.
输入模式
{
"type": "object",
"properties": {
"trace": {
"type": "array",
"items": {
"type": "object"
}
},
"pricing": {
"type": "object",
"nullable": true,
"properties": {
"in": {
"type": "number"
},
"out": {
"type": "number"
}
}
}
},
"required": [
"trace"
]
}社区
证据
最近观测
已验证未记录版本3 个工具
已验证未记录版本3 个工具