arthas-mcp-proxy
Aggregates multiple Arthas agents into a single MCP server, allowing Claude to diagnose any number of Java processes with just one MCP configuration.
README
arthas-mcp-proxy
Arthas 多 Agent 聚合代理 MCP。Claude 端只配 1 个 MCP,即可诊断任意多个已启动 arthas-agent 的 Java 进程。
原理
Claude (Code / Desktop) ──stdio──> server.py ──┬─ /mcp 转发(30 个官方工具)──> 各 arthas agent
│读 └─ /api 执行(list_agents/help/execute)
├─ agents.yaml (agent 清单 + tags)
└─ arthas_tools.json(官方 30 工具 schema,固化)
对外暴露 33 个工具,每个用 target 参数选目标 agent。新增服务只改 agents.yaml,Claude 端零改动。
工具
代理自有(走 /api):
list_agents(tag?)— 列出所有可用 target,可按 tag 筛选help(target, command?)— 查 arthas 命令的权威帮助/语法execute(target, command, timeout_ms?)— 兜底执行任意命令(危险命令走灰名单确认 / 黑名单拦截)
30 个官方工具(镜像 arthas 原生 /mcp,schema 与官方 100% 一致,走 /mcp 转发):
jvm dashboard memory thread sysprop sysenv perfcounter mbean vmoption vmtool options
sc sm jad getstatic classloader dump redefine retransform
trace watch stack monitor tt profiler
ognl mc heapdump viewfile stop
每个官方工具多一个
target参数,其余参数与官方完全相同。调用如thread(target="xxx服务", topN=3)。
工作机制
- 官方工具定义固化在
arthas_tools.json(启动不联网拉取)。MirrorTool子类读取它批量注册,运行时取出target→ 连该 agent 的/mcp→ 转调同名工具 → 原样返回。 - 因此工具的参数/描述始终与官方一致,且 arthas 升级后只需重新 dump 一次。
arthas 升级后刷新工具定义
.venv/bin/python -c "
import asyncio, json
from fastmcp import Client
from fastmcp.client.transports import StreamableHttpTransport
tp = StreamableHttpTransport('http://xx.xx.xx.xx:8563/mcp', headers={'Authorization':'Bearer arthas'})
async def main():
async with Client(tp) as c:
tools = await c.list_tools()
json.dump([{'name':t.name,'description':t.description,'inputSchema':t.inputSchema} for t in tools],
open('arthas_tools.json','w'), ensure_ascii=False, indent=2)
print('refreshed', len(tools))
asyncio.run(main())
"
配置(agents.yaml)
agents:
- name: "xxx服务"
url: http://xx.xx.xx.xx:8563
tags: ["测试", "IC", "biz"] # 打标签,便于 list_agents(tag=...) 批量定位
defaults:
username: arthas # 公共账号,所有 agent 共用(个别不同可在该 agent 下覆盖)
password: arthas
auth: basic # basic | bearer
exec_timeout_ms: 10000
confirm_commands: [stop, redefine, retransform, mc, reset] # 灰名单:执行前需用户确认
blocked_commands: [] # 黑名单:完全禁用、不暴露
配置路径默认本目录 agents.yaml,也可用环境变量 ARTHAS_AGENTS_CONFIG 指定。
安全控制:灰名单 / 黑名单
危险操作分两级,对 30 个官方工具和 execute 两条路径统一生效:
- 灰名单
confirm_commands:工具正常暴露,但带一个confirm参数。不带confirm=true调用会被拦下并返回警告,需在用户明确同意后带confirm=true重新调用才执行。默认含stop/redefine/retransform/mc/reset。 - 黑名单
blocked_commands:完全禁用——mirror 工具不暴露、execute 直接拒绝。默认空。
说明:原计划用 MCP elicitation 做弹窗确认,但实测 Claude 客户端在自动流程下会自动 decline、不弹窗,故改用更可靠的
confirm参数二次确认。改agents.yaml这两项即可调整,重启 Claude 生效。
安装
cd /path/to/arthas-mcp-proxy
uv venv && uv pip install fastmcp httpx pyyaml
接入 Claude
Claude Code(local scope):
claude mcp add arthas-proxy --scope local -- \
/path/to/arthas-mcp-proxy/.venv/bin/python /path/to/arthas-mcp-proxy/server.py
Claude Desktop(claude_desktop_config.json):
"arthas-proxy": {
"command": "/path/to/arthas-mcp-proxy/.venv/bin/python",
"args": ["/path/to/arthas-mcp-proxy/server.py"]
}
改动 server.py / agents.yaml 后需重启 Claude 生效。
本地冒烟测试
.venv/bin/python -c "
import asyncio, server
from fastmcp import Client
async def main():
async with Client(server.mcp) as c:
print('工具数:', len(await c.list_tools()))
r = await c.call_tool('jvm', {'target':'xxx服务'})
print(r.content[0].text[:120])
asyncio.run(main())
"
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.