arthas-mcp-proxy
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@arthas-mcp-proxycheck thread dump for payment-service with topN=5"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
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。
Related MCP server: MCP Server Proxy
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.pyClaude 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())
"This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Connectors
MCP server for building and testing AI agents with multi-model experimentation and insights.
Agent-native collaboration network: orchestrate a team of long-running agents from any MCP client.
The MCP server for Azure DevOps, bringing the power of Azure DevOps directly to your agents.
MCP server for progressive tool usage at any scale (see https://klavis.ai)
Related MCP Servers
- FlicenseNot gradedqualityDmaintenanceProxy that aggregates multiple MCP servers and presents them as a unified interface, allowing clients to access resources from multiple servers transparently.4-
- AlicenseNot gradedqualityDmaintenanceAggregates multiple MCP servers into a single endpoint, enabling LLM clients to access tools, resources, and prompts from various backends through one connection.18MIT
- AlicenseAqualityDmaintenanceAn MCP server that enables Claude to orchestrate multiple autonomous Claude Code agents working in parallel across different projects, with tools to dispatch, monitor, and manage their progress.6481MIT
- FlicenseNot gradedqualityCmaintenanceAggregates multiple MCP servers into a single unified endpoint with hot-plugging, multi-protocol support, and management via web and CLI.14-