1. Conduid
  2. Other
  3. Trace MCP
MCP server · Other

Trace MCP

Unclaimed other
40Fair

Scored 4 days ago · breakdown

About Trace MCP

Trace MCP is an MCP server published by OraingeWZC in the Other category.

The repository has 56 stars and 0 forks, with the last commit 4 months ago. Recent commit activity is the strongest signal that reported issues will get attention.

Install

From source
git clone https://github.com/OraingeWZC/Trace_mcp

This server has no ConduID identity, so agent calls to it are not receipted. Pin the version you install and review the source before granting it credentials.

Ask AI

✦

Ask AI about Trace MCP

Powered by Claude · Grounded in docs

I know everything about Trace MCP. Ask me about installation, configuration, usage, or troubleshooting.

Security checks

Static checks derived from the repository 3 days ago. Heuristics on the source, not an audit.

  • ✓README presentThe repository ships documentation.
  • !License declaredNo license file. Legally you have no right to use it in a product.
  • ✓Tests presentA test suite exists in the repository.
  • !Dependencies pinnedNo lockfile. Installs can pull different code over time.

2 checks have not been run yet.

README

Trace_mcp(MCP Server)

该目录提供一个基于 mcp.server.fastmcp.FastMCP 的 MCP Server,用于把 Trace_mcp/app/tools 下的故障诊断/根因分析脚本封装为可被 LLM 或其他 MCP Client 调用的工具。

目录结构(核心)

路径 作用 备注
Trace_mcp/app/main.py MCP Server 入口,注册对外 tools 只做“工具注册 + 参数打包”
Trace_mcp/app/handler.py 子进程执行器:把 tool 调用映射到具体脚本/模块并执行 统一日志、环境变量与参数转换
Trace_mcp/app/schemas.py MCP 侧请求/响应模型(Pydantic) ToolRequest / ScriptResult
Trace_mcp/app/tools/ 四个“故障处理工具”目录 每个目录内部是独立的脚本/模型代码
Trace_mcp/app/data/ 数据集处理脚本(按数据集划分) aiops25/、tianchi/

四个工具目录一览

tool_id(handler 使用) 目录 说明 stage -> 目标
aiops_svnd Trace_mcp/app/tools/trace_svnd_diag/ Trace + 拓扑/节点信息融合(SVND) preprocess->make_aiops_svnd.py / train->train_aiops_svnd.py / test->test_aiops_svnd.py / debug->debug_check_svnd.py
aiops_sv Trace_mcp/app/tools/trace_sv_diag/ 单 Trace 结构诊断(SV) preprocess->make_aiops_sv.py / train->train_aiops_sv.py / test->test_aiops_sv.py
tratoporca Trace_mcp/app/tools/TraTopoRca/ GTrace 风格根因分析(tracegnn) train->python -m tracegnn.models.gtrace.mymodel_main / test->python -m tracegnn.models.gtrace.mymodel_test
tracezly_rca Trace_mcp/app/tools/tracezly_rca/ 实验性脚本集合(含 gtrace train 与数据转换) preprocess->convert_trace_dataset.py / train->python -m tracegnn.models.gtrace.mymodel_main / test->evaluate_model.py

说明:上表的映射在 Trace_mcp/app/handler.py 的 get_tool_registry() 中集中维护。

MCP 暴露的工具列表(对外)

MCP tool 名称 用途 推荐场景
ping 连通性检查(不启动子进程) 验证 MCP 通信链路
list_trace_tools 列出 tool_id/stage 映射 调试/排查“脚本到底会跑哪个”
get_gtrace_config 读取 gtrace 的 config.py(白名单字段) 在训练/测试前查看当前可控参数与取值
update_gtrace_config 更新 gtrace 的 config.py(支持 dry-run + 自动备份) 用自然语言生成结构化 updates,安全落盘后再训练/测试
run_trace_tool 通用运行入口:tool_id + stage + args 统一入口,适合自动化/批量
check_env 环境诊断探针(走 aiops_svnd:debug) 排查子进程启动/依赖问题
preprocess_aiops_svnd / train_aiops_svnd / test_aiops_svnd SVND 三件套 Trace+拓扑融合诊断
preprocess_aiops_sv / train_aiops_sv / test_aiops_sv SV 三件套 单 Trace 结构诊断
train_tracerca / test_tracerca TraTopoRca(GTrace) 训练/评估 根因分析实验
preprocess_tracezly_rca / train_tracezly_rca / test_tracezly_rca tracezly_rca 相关入口 实验性脚本

运行方式

方式 命令 说明
直接启动 python -m app.main 在 Trace_mcp/ 目录下执行
Inspector 启动 npx @modelcontextprotocol/inspector python -m app.main 适合本地调试/查看 schema

TraTopoRca / tracezly_rca 的启动方式与配置说明(重要)

你提到的两点这里做显式说明:

工具目录 训练命令(在该目录下运行) 测试命令(在该目录下运行) 参数控制方式
Trace_mcp/app/tools/TraTopoRca/ python -m tracegnn.models.gtrace.mymodel_main python -m tracegnn.models.gtrace.mymodel_test 主要由 Trace_mcp/app/tools/TraTopoRca/tracegnn/models/gtrace/config.py 控制;mymodel_test 额外支持命令行覆盖(如 --model/--report-dir/--limit 等)
Trace_mcp/app/tools/tracezly_rca/ python -m tracegnn.models.gtrace.mymodel_main (无标准 mymodel_test.py)可用 python evaluate_model.py 主要由 Trace_mcp/app/tools/tracezly_rca/tracegnn/models/gtrace/config.py 控制

调用示例(通用入口 run_trace_tool)

{
  "tool_id": "aiops_svnd",
  "stage": "test",
  "args": {
    "model_path": "dataset/aiops_svnd/1019/aiops_nodectx_multihead.pt",
    "data_root": "dataset/aiops_svnd",
    "batch_size": 128,
    "device": "cuda",
    "limit": 50
  },
  "timeout_sec": 600,
  "tail_lines": 200
}

日志与排查

项 位置 说明
子进程完整日志 各工具目录下的 mcp_<tool_id>_<stage>_<timestamp>.log handler.py 自动落盘
MCP 返回日志 ScriptResult.log 仅返回 tail,避免响应过大

数据处理(app/data)

目录 数据集 说明
Trace_mcp/app/data/aiops25/ AIOps25 该目录下是按比赛/数据集定制的处理脚本(如 trace 切分、groundtruth 对齐等)
Trace_mcp/app/data/tianchi/ 天池 包含数据/脚本/src/tools 等子目录,适合复用既有 pipeline

README fetched from the repository 3 days ago. The original is authoritative.

Questions

About Trace MCP

How do I install Trace MCP?

No package has been published for this server, so install it from source: clone the repository and follow its README.

Is Trace MCP safe to use with an AI agent?

Its trust score is 40 out of 100 (fair). It passes 2 of the 4 static security checks run so far; the failures are listed above. It has no ConduID identity yet, so agent calls to it are not receipted.

Is Trace MCP still maintained?

Yes — the last commit was 4 months ago. The repository has 56 stars and 0 forks.