Arthas MCP Proxy
Enables dynamic SSH connection to servers for real-time JVM diagnostics using Arthas, allowing operations like thread dump, heap inspection, and method watching through MCP tools.
README
Arthas MCP Proxy
MCP Server for JVM diagnostics via SSH + Arthas.
Provides 26+ diagnostic tools through the Model Context Protocol (MCP), enabling AI assistants to perform thread dumps, heap analysis, method tracing, CPU profiling, and more on remote Java processes.
Features
- Full Arthas command suite —
thread,trace,watch,heapdump,profiler,jad, etc. - Multi-target SSH — connect to multiple JVM hosts from a single server
- Cross-user diagnosis — automatically uses
sudo -u <owner>when SSH user != process owner - Concurrent-safe — per-PID attach locks + three-level reuse (cache → detect → attach)
- SSE & stdio transports — works with Cursor, Claude Desktop, and other MCP clients
Quick Start
Docker (recommended)
docker run -p 8000:8000 ghcr.io/narcissux/arthas-mcp-proxy:latest
From source
git clone https://github.com/narcissux/arthas-mcp-proxy.git
cd arthas-mcp-proxy
pip install -e ".[dev]"
python -m arthas_mcp_proxy --transport sse --port 8000
Cursor / MCP Client configuration
{
"mcpServers": {
"arthas": {
"url": "http://localhost:8000/sse"
}
}
}
Available Tools
| Tool | Purpose |
|---|---|
connect_ssh |
Establish SSH connection to target server |
list_java_processes |
List Java processes with Arthas status |
thread_dump |
Thread dump (top N by CPU) |
heap_info |
Memory dashboard |
watch_method |
Watch method params/return values |
exec_command |
Universal Arthas command executor (26+ commands) |
install_arthas |
Install Arthas on target server |
disconnect_ssh |
Disconnect and release resources |
Development
Running tests
# Unit tests only (mocked, no external dependencies)
pytest tests/ --ignore=tests/integration/
# Integration tests with auto-managed Docker target (recommended)
pytest tests/integration/ -m integration -v --docker-target
# Integration tests against a remote target (env vars required)
export TEST_SSH_HOST=your-server
export TEST_SSH_USER=your-username
export TEST_SSH_PASSWORD=your-password
pytest tests/integration/ -m integration -v
# Manual two-step Docker target
docker compose -f docker-compose.test.yml up --build -d
export TEST_SSH_HOST=localhost TEST_SSH_USER=testuser TEST_SSH_PASSWORD=testpass
pytest tests/integration/ -m integration -v
docker compose -f docker-compose.test.yml down --volumes
Code quality
# Install dev dependencies
pip install -e ".[dev]"
# Run lint
ruff check src/ tests/
ruff format --check src/ tests/
# Run type check
mypy src/arthas_mcp_proxy
# Run tests
pytest -v
# Run with coverage
pytest --cov=arthas_mcp_proxy --cov-report=html
Project Structure
.
├── src/
│ └── arthas_mcp_proxy/
│ ├── __init__.py # Package init
│ ├── __main__.py # python -m arthas_mcp_proxy
│ ├── server.py # MCP server & tools
│ ├── arthas_client.py # Arthas attach & command execution
│ ├── ssh_pool.py # SSH connection pool
│ └── decorators.py # @require_session and other decorators
├── tests/
│ ├── conftest.py # Shared fixtures (mock SSH session, state cleanup)
│ ├── test_decorators.py # @require_session tests
│ ├── test_arthas_client.py # Concurrency & logic tests
│ ├── test_ssh_pool.py # Connection pool tests
│ └── integration/
│ ├── conftest.py # Integration env validation & Docker check
│ └── test_real_jvm.py # Real JVM diagnostic tests via SSH
├── pyproject.toml # Project config, deps, tool settings
├── entrypoint.sh # Test target container startup script
├── Dockerfile
├── docker-compose.yml # Production deployment
├── docker-compose.test.yml # Test infrastructure (SSH + Java container)
├── Dockerfile.test-target # Test target image (Java + math-game.jar)
└── README.md
Test categories
| Category | Command | Requirements |
|---|---|---|
| Unit tests | pytest tests/ --ignore=tests/integration/ |
None (fully mocked) |
| Integration (remote) | pytest tests/integration/ -m integration |
SSH target with Java |
| Integration (Docker) | docker compose -f docker-compose.test.yml up + pytest |
Docker daemon |
Environment variables for integration tests
| Variable | Required | Default | Description |
|---|---|---|---|
TEST_SSH_HOST |
Yes | — | Target hostname or IP |
TEST_SSH_USER |
Yes | — | SSH username |
TEST_SSH_PASSWORD |
Yes | — | SSH password |
TEST_SSH_PORT |
No | 22 | SSH port |
TEST_TARGET_PID |
No | auto | Specific PID to diagnose |
Security: Never commit credentials. Use environment variables or a .env
file (ignored by .gitignore).
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。