MCP Pprof

mcp-pprof

English | 中文

English

A Model Context Protocol (MCP) Server for Go pprof data analysis. This server wraps go tool pprof to provide AI-accessible tools for performance profiling and analysis.

Overview

mcp-pprof enables AI agents to:

• Parse and analyze Go pprof files (CPU, heap, block, mutex, goroutine profiles)
• Generate SVG flamegraphs
• Identify performance bottlenecks and hotspots
• Compare different profile files
• Provide optimization suggestions

Features

• MCP Protocol Support: Fully compatible with MCP protocol for seamless AI integration
• Multiple Transport Modes: Supports both stdio (local) and HTTP (mcp-remote) modes
• Comprehensive Tools:

- parse_profile - Parse pprof files and return structured data - top_functions - Get top N hot functions - generate_svg - Generate SVG flamegraphs - analyze_performance - Deep performance analysis with suggestions - compare_profiles - Compare two profile files - list_callers - View function call relationships

Installation

Build from source

# Clone the repository
git clone https://github.com/gwork1883/mcp-pprof.git
cd mcp-pprof

# Build binaries
go build -o build/mcp-pprof ./cmd/mcp-pprof
go build -o build/mcp-pprof-server ./cmd/mcp-pprof-server

# Or use Makefile
make build

Requirements

• Go 1.21 or higher
• go tool pprof (included with Go)

Usage

Local Mode (stdio)

Add to your MCP configuration:

{
  "mcpServers": {
    "pprof": {
      "type": "stdio",
      "command": "/path/to/mcp-pprof",
      "args": [],
      "disabled": false,
      "autoApprove": [
        "parse_profile",
        "top_functions",
        "generate_svg",
        "analyze_performance",
        "compare_profiles",
        "list_callers"
      ]
    }
  }
}

Remote Mode (mcp-remote)

1. Start HTTP Server

./build/mcp-pprof-server -port 8080

2. Configure mcp-remote

{
  "mcpServers": {
    "pprof": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote@latest",
        "http://localhost:8080/mcp",
        "--allow-http",
        "--reconnect",
        "--reconnect-delay",
        "5000"
      ],
      "disabled": false,
      "autoApprove": [
        "parse_profile",
        "top_functions",
        "generate_svg",
        "analyze_performance",
        "compare_profiles",
        "list_callers"
      ]
    }
  }
}

Tools

Example Usage with AI

Please analyze the CPU profile at path/to/profile.prof

Generate a flamegraph for the heap profile

Compare the before.prof and after.prof profiles

Development

# Build
make build

# Run stdio mode
make run-stdio

# Run HTTP server
make run-server

# Test
make test

Architecture

mcp-pprof/
├── cmd/
│   ├── mcp-pprof/           # stdio mode entry
│   └── mcp-pprof-server/    # HTTP server entry
├── internal/
│   ├── mcp/                 # MCP protocol implementation
│   ├── pprof/               # go tool pprof wrapper
│   └── tools/               # Tool handlers
├── pkg/
│   └── protocol/            # MCP protocol types
└── docs/                    # Documentation

License

MIT License

中文

用于 Go pprof 数据分析的 Model Context Protocol (MCP) Server。该服务器封装了 go tool pprof，为 AI 提供性能分析工具。

概述

mcp-pprof 使 AI 智能体能够：

• 解析和分析 Go pprof 文件（CPU、heap、block、mutex、goroutine profiles）
• 生成 SVG 火焰图
• 识别性能瓶颈和热点
• 对比不同的 profile 文件
• 提供优化建议

功能特性

• MCP 协议支持：完全兼容 MCP 协议，无缝集成 AI
• 多种传输模式：支持 stdio（本地）和 HTTP（mcp-remote）两种模式
• 丰富的工具集：

- parse_profile - 解析 pprof 文件并返回结构化数据 - top_functions - 获取 Top N 热点函数 - generate_svg - 生成 SVG 火焰图 - analyze_performance - 深度性能分析与优化建议 - compare_profiles - 对比两个 profile 文件 - list_callers - 查看函数调用关系

安装

从源代码构建

# 克隆仓库
git clone https://github.com/gwork1883/mcp-pprof.git
cd mcp-pprof

# 构建二进制文件
go build -o build/mcp-pprof ./cmd/mcp-pprof
go build -o build/mcp-pprof-server ./cmd/mcp-pprof-server

# 或使用 Makefile
make build

系统要求

• Go 1.21 或更高版本
• go tool pprof（Go 自带）

使用方式

本地模式 (stdio)

将以下配置添加到您的 MCP 配置中：

{
  "mcpServers": {
    "pprof": {
      "type": "stdio",
      "command": "/path/to/mcp-pprof",
      "args": [],
      "disabled": false,
      "autoApprove": [
        "parse_profile",
        "top_functions",
        "generate_svg",
        "analyze_performance",
        "compare_profiles",
        "list_callers"
      ]
    }
  }
}

远程模式 (mcp-remote)

1. 启动 HTTP 服务器

./build/mcp-pprof-server -port 8080

2. 配置 mcp-remote

{
  "mcpServers": {
    "pprof": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote@latest",
        "http://localhost:8080/mcp",
        "--allow-http",
        "--reconnect",
        "--reconnect-delay",
        "5000"
      ],
      "disabled": false,
      "autoApprove": [
        "parse_profile",
        "top_functions",
        "generate_svg",
        "analyze_performance",
        "compare_profiles",
        "list_callers"
      ]
    }
  }
}

工具

AI 使用示例

请分析这个 CPU profile 文件，告诉我哪些函数占用最多时间

为这个 heap profile 生成火焰图

对比优化前后的两个 profile 文件，找出性能改进点

开发

# 构建
make build

# 运行 stdio 模式
make run-stdio

# 运行 HTTP 服务器
make run-server

# 运行测试
make test

项目架构

mcp-pprof/
├── cmd/
│   ├── mcp-pprof/           # stdio 模式入口
│   └── mcp-pprof-server/    # HTTP 服务器入口
├── internal/
│   ├── mcp/                 # MCP 协议实现
│   ├── pprof/               # go tool pprof 包装器
│   └── tools/               # 工具处理器
├── pkg/
│   └── protocol/            # MCP 协议类型定义
└── docs/                    # 文档

许可证

MIT License