Table of Contents

Arthas MCP Server

Overview

Arthas MCP Server is an experimental module of Arthas that implements a server based on the MCP (Model Context Protocol). This module provides a unified JSON-RPC 2.0 interface through HTTP/Netty, enabling AI assistants to execute Arthas diagnostic commands via tool calls.

MCP (Model Context Protocol) is a standardized protocol proposed by Anthropic for connecting AI assistants with various tools and data sources. Through Arthas MCP Server, AI assistants can naturally execute Java application diagnostic tasks, significantly improving development and operations efficiency.

Key Features

• AI-Native Integration: Supports mainstream AI clients (Claude Desktop, Cherry Studio, Cline, etc.)
• Standardized Protocol: Full implementation of MCP protocol specification (version 2025-03-26), supporting Streamable Http transport
• 26 Diagnostic Tools: Covers core functionalities including JVM monitoring, class loading, method tracing, etc.
• Security Authentication: Supports Bearer Token authentication mechanism

Supported Diagnostic Tools

Arthas MCP Server integrates 26 core diagnostic tools, categorized by functionality:

JVM-Related Tools (12)

Tool	Description
dashboard	Real-time JVM/application dashboard with customizable refresh interval and count control
heapdump	Generate JVM heap dump file, supports --live option to export only live objects
jvm	View current JVM detailed information
mbean	View or monitor MBean attributes, supports real-time refresh and pattern matching
memory	View JVM memory information
thread	View thread information and stack traces, supports finding blocked and busiest threads
sysprop	View or modify system properties, supports dynamic JVM system property modification
sysenv	View system environment variables
vmoption	View or update VM options, supports dynamic JVM parameter adjustment
perfcounter	View Perf Counter information, displays JVM performance counters
vmtool	VM tool collection, supports forced GC, instance retrieval, thread interruption, etc.
getstatic	View static field values of a class
ognl	Execute OGNL expressions, dynamically invoke methods and access fields

Class/ClassLoader Tools (8)

Tool	Description
sc	View loaded class information in JVM, supports detailed info and statistics
sm	View method information of loaded classes, displays method signatures and modifiers
jad	Decompile loaded class source code, converts bytecode to Java code
classloader	ClassLoader diagnostic tool, view classloader statistics, hierarchy tree, URLs
mc	Memory compiler, compiles Java source code to bytecode files
redefine	Redefine classes, load external class files to redefine classes in JVM
retransform	Retransform classes, triggers class retransformation and bytecode enhancement
dump	Dump actual running class bytecode from JVM to specified directory

Monitoring and Diagnostic Tools (6)

Tool	Description
monitor	Monitor invocation of specified methods in real-time
stack	Output call path of current method, helps analyze method call chains
trace	Trace internal method call paths, output time cost for each node
tt	Time tunnel for method execution data, records parameters and return values for each invocation
watch	Observe method invocations including parameters, return values, and exceptions

Quick Start

1. Configure MCP Service

Enable MCP service in arthas.properties configuration file:

# MCP (Model Context Protocol) configuration
arthas.mcpEndpoint=/mcp

2. Start Application

Start Arthas or Java application with Arthas normally. By default, MCP service will be exposed on HTTP port 8563.

Verify MCP service is running:

curl http://localhost:8563/mcp

If MCP protocol information is returned, the service has started successfully.

3. Configure AI Client

Configuration examples for mainstream AI clients:

Cherry Studio / Cline

Add MCP server configuration in settings:

{
  "mcpServers": {
    "arthas-mcp": {
      "type": "streamableHttp",
      "url": "http://localhost:8563/mcp"
    }
  }
}

Configuration

Arthas Configuration

Property	Description	Default
arthas.mcpEndpoint	MCP service access path	None (requires manual configuration)
arthas.httpPort	HTTP service port	8563
arthas.password	Authentication password (when authentication enabled)	None

Authentication Configuration

When arthas.password is configured in the configuration file, MCP Server will automatically enable authentication. In this case, you need to add an authentication header in the AI client configuration, with the Bearer Token set to the configured password value.

Configuration file example (arthas.properties):

arthas.password=your-secure-password

AI client configuration example:

{
  "mcpServers": {
    "arthas-mcp": {
      "type": "streamableHttp",
      "url": "http://localhost:8563/mcp",
      "headers": {
        "Authorization": "Bearer your-secure-password"
      }
    }
  }
}

WARNING

Note: The token in the Authorization header must exactly match the value configured in arthas.password.

Feedback & Contribution

TIP

Arthas MCP Server is an experimental feature. Feedback and suggestions are welcome!

• Issue Report: GitHub Issues
• Feature Request: GitHub Discussions
• Contribute: Contributing Guide