gdb-mcp
Enables AI assistants to control GDB debugger via MCP protocol for local and remote debugging, supporting CTF Pwn, crash analysis, and ELF inspection.
README
gdb-mcp 是一款高权限模型上下文协议(MCP)服务端,可让 Codex 通过 GDB/MI 接口与 pygdbmi 库控制 GDB 调试器。
该项目主要支持的部署架构:
Windows 上的 Codex
-> Windows 上的 gdb-mcp MCP 服务端
-> Windows 本地 GDB
-> target remote / target extended-remote
-> linux 虚拟机中的 gdbserver
-> 虚拟机内的目标程序
本服务端专为本地授权调试、CTF Pwn 解题、崩溃分析、Core Dump 分析、漏洞利用复现、ELF 文件分析场景设计。高风险 GDB 命令不会被永久禁用,首次执行会返回警告,调用方在重试时传入 confirm=true 即可运行。
工具列表
gdb_session:本地 GDB 会话的启动、停止、重启与状态查询gdb_load:加载本地二进制文件、Core 文件、符号文件与启动参数gdb_exec:通用 GDB CLI 命令执行入口gdb_mi:原生 GDB/MI 命令执行入口gdb_remote:配置并连接虚拟机端gdbservergdb_context:查看寄存器、栈、反汇编、调用栈、断点、内存映射、共享库gdb_memory:内存读取、写入、搜索与转储gdb_register:寄存器读写gdb_breakpoint:普通断点、硬件断点、临时断点、观察点gdb_run_control:运行、继续、单步步入、步过、指令级单步、执行完函数、中断、杀死程序、重启gdb_analyze:崩溃与漏洞利用可行性分析gdb_elf:安全检查(checksec)、ELF 头、节、段、符号、GOT/PLT、重定位、字符串gdb_pwndbg:兼容执行 pwndbg/gef/peda 命令
所有工具返回统一 JSON 结构:
{
"ok": true,
"tool": "gdb_exec",
"action": "exec",
"risk_level": "low",
"need_confirm": false,
"executed_with_risk": false,
"warning": null,
"data": {},
"stdout": "",
"stderr": "",
"raw": {},
"error": null
}
安装
git clone https://github.com/traver88/gdb-mcp.git
本项目支持的运行环境为 Python 3.12。
Windows 环境:Codex + Windows gdb-mcp + 虚拟机 gdbserver
在项目目录安装 Python隔离环境
python --version
python -m venv .venv
.\.venv\Scripts\activate
pip install -e .
Windows:安装 GDB
方式 A:MSYS2 MinGW64 GDB
- 安装 MSYS2
- 打开 MSYS2 MinGW64 终端
- 执行:
pacman -Syu # 之后重启 MSYS2
pacman -S gdb gdb-multiarch
- 将
C:\msys64\mingw64\bin加入 WindowsPATH - 验证:
gdb --version
gdb-multiarch --version
运行smoke_test测试
gdb-mcp项目目录下:
.\.venv\Scripts\activate
python tests\smoke_test.py
成功会输出:
smoke test passed
smoke_test.py测试内容:
- 启动 Windows 本地 GDB
- 加载
examples/hello.c编译的测试程序 - 在
main下断点 - 运行程序
- 读取寄存器
- 读取栈
- 反汇编
- 关闭 GDB
虚拟机(Ubuntu/Kali)配置
安装 gdbserver
sudo apt update
sudo apt install -y gdbserver gdb gcc make binutils file
启动目标程序
cd xxx/xxx/
gdbserver 0.0.0.0:1234 ./pwn
Windows 端连通性测试
powershell:
ping 192.168.56.101
Test-NetConnection 192.168.56.101 -Port 1234
Codex 配置 config.toml
[mcp_servers.gdb-mcp]
command = "D:\\gdb-mcp\\.venv\\Scripts\\python.exe"
args = ["D:\\gdb-mcp\\server.py"]
command:指向启动 MCP 服务端的 Python 程序args:指向server.py服务端入口
Codex 使用示例
需求示例:
请使用 gdb-mcp:
- 启动 GDB
- 加载本地符号文件 E:/ctf/pwn/pwn
- 连接 192.168.56.101:1234 的 gdbserver
- 在 main 下断点
- continue
- 显示寄存器、栈、RIP 附近反汇编、backtrace
等价 MCP 调用:
gdb_session(action="start")
gdb_remote(
action="connect",
host="192.168.56.101",
port=1234,
mode="remote",
local_binary="E:/ctf/pwn/pwn",
confirm=true
)
gdb_breakpoint(action="add", location="main")
gdb_run_control(action="continue")
gdb_register(action="read_all")
gdb_memory(action="read", address="$rsp", size=160)
gdb_exec(command="x/20i $rip")
gdb_exec(command="bt")
gdb_context(depth=20)
等价 GDB 命令:
file "E:/ctf/pwn/pwn"
target remote 192.168.56.101:1234
b main
c
info registers
x/20gx $rsp
x/20i $rip
bt
扩展远程模式(Extended remote)
虚拟机:
gdbserver --multi 0.0.0.0:1234
MCP 调用:
gdb_remote(
action="connect",
host="192.168.56.101",
port=1234,
mode="extended-remote",
local_binary="E:/ctf/pwn/pwn",
remote_binary="/xxx/pwn",
confirm=true
)
gdb_breakpoint(action="add", location="main")
gdb_run_control(action="run")
gdb_context(depth=20)
内部 GDB 命令:
file "E:/ctf/pwn/pwn"
set remote exec-file /xxx/pwn
target extended-remote 192.168.56.101:1234
常见问题
- Windows 无法 ping 通虚拟机:检查虚拟机 IP 与网卡模式
- 1234 端口无法连接:确认
gdbserver正在运行 gdbserver仅监听本地:使用gdbserver 0.0.0.0:1234 ./pwn- Windows 防火墙拦截:放行 GDB 或对应端口
- 虚拟机 NAT 模式无法访问:改用仅主机 / 桥接网卡或配置端口转发
local_binary与remote_binary混淆:前者为 Windows 路径,后者为 Linux 路径- Windows GDB 无法解析 Linux ELF:使用对应架构的多架构 GDB
- 远程模式下
info proc mappings不可用:gdb_context会标记为不可用而非崩溃 - 远程 libc 符号缺失:使用
gdb_remote(action="set_sysroot", ...)或set_solib_search_path
风险操作确认
需确认的高风险命令示例:
target remote HOST:PORTtarget extended-remote HOST:PORTdisconnectdetachset remote exec-file PATHset sysroot PATHset solib-search-path PATHshell ...source ...python ...dump memory ...restore ...maintenance ...
首次调用:
gdb_exec(command="target remote 192.168.56.101:1234")
服务端返回 need_confirm=true,带确认重试:
gdb_exec(command="target remote 192.168.56.101:1234", confirm=true)
执行后返回中 executed_with_risk=true。
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.