# mcp-debugger — DAP Debugging Tools for Agents

> Add step-through debugging to AI agents via an MCP server built on the Debug Adapter Protocol (DAP). Works across languages; run with npm or Docker.

## Install

Merge the JSON below into your `.mcp.json`:

## Quick Use

1. Install / run:
   ```bash
   npx @debugmcp/mcp-debugger --help
   ```
2. Start / smoke test:
   ```bash
   npm install -g @debugmcp/mcp-debugger && mcp-debugger --help
   ```
3. Verify:
   - Start the MCP server and call `list_supported_languages`, then create one debug session and set a breakpoint.

## Intro

Add step-through debugging to AI agents via an MCP server built on the Debug Adapter Protocol (DAP). Works across languages; run with npm or Docker.

- **Best for:** AI coding agents that need deterministic debugging actions (breakpoints, stepping, stack/vars) instead of log guessing
- **Works with:** MCP clients + DAP-based debuggers; multi-language projects (see repo for adapters)
- **Setup time:** 10 minutes

## How to Use It Well

- Setup time ~10 minutes (npm + connect from your MCP client)
- GitHub stars + forks (verified): see Source & Thanks
- One MCP tool call can replace multiple “please step/print/inspect” prompt turns

If your agent’s debugging loop is “edit → run → read logs → guess”, you waste tokens and time. With mcp-debugger, you can make debugging state explicit: set one breakpoint, step a few frames, and inspect locals/stack via structured calls.

### FAQ

**Q: What does it actually expose?**
A: A set of MCP tools that map to DAP-style debugging actions like create session, breakpoints, stepping, stack and variables.

**Q: Do I need an IDE UI?**
A: No. Agents can drive debugging headlessly by calling MCP tools; your IDE is optional.

**Q: Is it safe for production?**
A: Use it in dev/CI environments. Limit what files/processes the debugger can touch and run it with least privilege.

## Source & Thanks

> Source: https://github.com/debugmcp/mcp-debugger
> License: MIT
> GitHub stars: 105 · forks: 15

---

<!-- ZH -->

## 快速使用

1. 安装 / 运行：
   ```bash
   npx @debugmcp/mcp-debugger --help
   ```
2. 启动 / 冒烟测试：
   ```bash
   npm install -g @debugmcp/mcp-debugger && mcp-debugger --help
   ```
3. 验证：
   - Start the MCP server and call `list_supported_languages`, then create one debug session and set a breakpoint.

## 简介

通过 MCP server 把 Debug Adapter Protocol（DAP）的断点/单步/变量查看等能力暴露为结构化工具调用，让 AI agent 能在多语言项目里做可重复的调试排障。

- **适合谁：** 希望把“断点/单步/堆栈/变量”变成可控工具调用，而不是靠日志猜测的 AI 编程代理
- **可搭配：** 支持 MCP 的客户端 + 基于 DAP 的调试器；多语言工程（适配器以仓库为准）
- **准备时间：** 10 分钟

## 实战建议

- 接入时间约 10 分钟（npm 安装 + 在 MCP 客户端里连接）
- GitHub stars + forks（已核验）：见「来源与感谢」
- 一次工具调用可替代多轮“单步/打印/检查变量”的对话回合

如果你的 agent 调试循环是“改代码→跑→看日志→猜原因”，通常会浪费大量 token 和时间。mcp-debugger 的价值在于把调试状态显式化：下一个断点、走几步、查看局部变量/堆栈，都变成结构化的可重复工具调用。

### FAQ

**它到底暴露了什么能力？**
A: 它把 DAP 风格的调试动作封装成 MCP 工具：创建会话、下断点、单步、查看堆栈与变量等。

**必须配 IDE UI 吗？**
A: 不需要。Agent 通过 MCP 工具就能无界面调试；IDE 只是可选的可视化入口。

**适合放到生产环境吗？**
A: 更适合开发/CI。建议最小权限运行，限制可调试的进程与可访问的路径。

## 来源与感谢

> Source: https://github.com/debugmcp/mcp-debugger
> License: MIT
> GitHub stars: 105 · forks: 15


---
Source: https://tokrepo.com/en/workflows/mcp-debugger-dap-debugging-tools-for-agents
Author: MCP Hub