做什么
通过 Chrome DevTools 协议进行综合节点.js 调试的 MCP 服务器,其中包含断点、踩踏、可变检查和源图
技能库 智客分类:Agent 工作流 devtools-debugger-mcp-nodejs
通过 Chrome DevTools 协议进行综合节点.js 调试的 MCP 服务器,其中包含断点、踩踏、可变检查和源图
官方网址:skills.sh
先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。
通过 Chrome DevTools 协议进行综合节点.js 调试的 MCP 服务器,其中包含断点、踩踏、可变检查和源图
官方 description 未单独写出 Use when。按规范,代理会在用户任务与这段 description 的关键词匹配时激活本技能。
按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:devtools-debugger-mcp-nodejs、What It Does、Installation、Configuration、MCP Settings、Output Format。 其中含规范建议的小节:分步指令。
文件分析:这是一份仅含 SKILL.md 的指令型技能,代理激活后整份正文进入上下文。 正文约 787 行,超过规范建议的 500 行,激活时仍会整份载入。
MCP server for comprehensive Node.js debugging via Chrome DevTools Protocol with breakpoints, stepping, variable inspection, and source maps
devtools-debugger-mcp-nodejsWhat It DoesInstallationConfigurationMCP SettingsOutput FormatCore Debugging Workflow1. Start Debug Session2. Set Breakpoints3. Exception Breakpoints4. Resume and Step5. Inspect Variables and Scopes
来源分类:skills.sh agent-skill
namedevtools-debugger-mcp-nodejsdescription具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗
先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。
复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。
把 Agent Skill「devtools-debugger-mcp-nodejs」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-08115f67f72415c9-Devtools-%E8%B0%83%E8%AF%95%E5%99%A8-Mcp-%E8%8A%82%E7%82%B9.html 请存为 .cursor/skills/devtools-debugger-mcp-nodejs/SKILL.md 或 .claude/skills/devtools-debugger-mcp-nodejs/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。
需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。
npx skills add 'https://github.com/reason-machines/devtools-skills' --list
npx skills add 'https://github.com/reason-machines/devtools-skills' --skill 'devtools-debugger-mcp-nodejs'
CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。
name: devtools-debugger-mcp-nodejs description: MCP server for comprehensive Node.js debugging via Chrome DevTools Protocol with breakpoints, stepping, variable inspection, and source maps triggers:
Skill by ara.so — Devtools Skills collection.
An MCP (Model Context Protocol) server that exposes comprehensive Node.js debugging capabilities through the Chrome DevTools Protocol. Enables AI assistants to set breakpoints, step through code, inspect variables, evaluate expressions, analyze call stacks, and work with source maps — all programmatically.
This MCP server launches Node.js applications with the built-in inspector (--inspect-brk=0), connects via WebSocket to the Chrome DevTools Protocol, and exposes debugging operations as MCP tools. It handles:
this), drill into object propertiesnpm install devtools-debugger-mcp
Or globally:
npm install -g devtools-debugger-mcp
Add to your MCP client configuration (e.g., Claude Desktop's claude_desktop_config.json):
{
"mcpServers": {
"devtools-debugger": {
"command": "node",
"args": ["/path/to/devtools-debugger-mcp/dist/index.js"]
}
}
}
Or if installed globally:
{
"mcpServers": {
"devtools-debugger": {
"command": "devtools-debugger-mcp"
}
}
}
Set default response format (text, json, or both):
// Via set_output_format tool (defaults to 'text')
{
"tool": "set_output_format",
"params": { "format": "json" }
}
Individual tools can override with their own format parameter.
// Launch Node.js script with inspector
{
"tool": "start_node_debug",
"params": {
"scriptPath": "/absolute/path/to/app.js",
"format": "text" // optional: 'text' | 'json' | 'both'
}
}
Returns initial pause at first line with pauseId and top frame info.
With arguments and environment:
{
"tool": "start_node_debug",
"params": {
"scriptPath": "/path/to/server.js",
"args": ["--port", "3000"],
"env": {
"NODE_ENV": "development",
"DEBUG": "*"
}
}
}
File path + line (1-based):
{
"tool": "set_breakpoint",
"params": {
"filePath": "/path/to/app.js",
"line": 42
}
}
Conditional breakpoint:
{
"tool": "set_breakpoint_condition",
"params": {
"filePath": "/path/to/users.js",
"line": 15,
"condition": "user.age > 18"
}
}
URL regex breakpoint (for modules/packages):
{
"tool": "set_breakpoint_condition",
"params": {
"urlRegex": ".*express.*",
"line": 100,
"condition": "req.method === 'POST'"
}
}
Logpoint (logs message without pausing):
{
"tool": "add_logpoint",
"params": {
"filePath": "/path/to/api.js",
"line": 28,
"message": "Request received: {req.url}"
}
}
{
"tool": "set_exception_breakpoints",
"params": {
"state": "uncaught" // 'none' | 'uncaught' | 'all'
}
}
Resume to next breakpoint:
{
"tool": "resume_execution",
"params": {
"includeScopes": true,
"includeStack": true,
"includeConsole": true,
"format": "text"
}
}
Step over current line:
{
"tool": "step_over",
"params": {
"includeScopes": true,
"includeConsole": true
}
}
Step into function:
{
"tool": "step_into",
"params": {
"includeStack": true
}
}
Step out of current function:
{
"tool": "step_out",
"params": {
"includeScopes": true
}
}
Continue to specific location:
{
"tool": "continue_to_location",
"params": {
"filePath": "/path/to/app.js",
"line": 55,
"column": 10 // optional
}
}
Current scope (locals, closures, this):
{
"tool": "inspect_scopes",
"params": {
"maxProps": 20, // max properties per object
"pauseId": "pause123", // optional, defaults to current
"frameIndex": 0, // optional, defaults to 0 (top frame)
"includeThisPreview": true,
"format": "text"
}
}
Drill into object properties:
// First get objectId from inspect_scopes or evaluate_expression
{
"tool": "get_object_properties",
"params": {
"objectId": "object:123",
"maxProps": 50
}
}
{
"tool": "evaluate_expression",
"params": {
"expr": "user.profile.email",
"pauseId": "pause123", // optional
"frameIndex": 0, // optional, which frame to eval in
"returnByValue": true, // optional, serialize result
"format": "json"
}
}
Evaluate with side effects:
{
"tool": "evaluate_expression",
"params": {
"expr": "items.push({ id: 5, name: 'test' }); items.length"
}
}
{
"tool": "list_call_stack",
"params": {
"depth": 10, // optional, max frames
"pauseId": "pause123", // optional
"includeThis": true, // optional, include 'this' preview
"format": "text"
}
}
{
"tool": "get_pause_info",
"params": {
"pauseId": "pause123", // optional, defaults to current
"format": "text"
}
}
Returns pause reason (breakpoint, exception, step, etc.) and location.
{
"tool": "read_console",
"params": {
"format": "text"
}
}
Retrieves console messages buffered since last step/resume. Console is also auto-included when includeConsole: true on step/resume tools.
{
"tool": "stop_debug_session"
}
Kills the Node.js process and cleans up CDP connection.
{
"tool": "list_scripts"
}
Returns all scripts loaded by Node.js (app files, node_modules, builtins).
// By scriptId
{
"tool": "get_script_source",
"params": {
"scriptId": "42"
}
}
// By URL
{
"tool": "get_script_source",
"params": {
"url": "file:///path/to/app.js"
}
}
{
"tool": "blackbox_scripts",
"params": {
"patterns": [
"node_modules/express/*",
"internal/*"
]
}
}
Frames matching these patterns won't pause during step-into.
Re-execute a specific call frame:
{
"tool": "restart_frame",
"params": {
"frameIndex": 2, // which frame to restart (0 = top)
"pauseId": "pause123", // optional
"format": "text"
}
}
Source maps are automatically detected and used. Just launch your compiled JS:
{
"tool": "start_node_debug",
"params": {
"scriptPath": "/path/to/dist/app.js"
}
}
// Set breakpoints using original .ts file paths
{
"tool": "set_breakpoint",
"params": {
"filePath": "/path/to/src/app.ts",
"line": 42
}
}
// 1. Start session
start_node_debug({ scriptPath: "/path/to/app.js" })
// 2. Set conditional breakpoint
set_breakpoint_condition({
filePath: "/path/to/app.js",
line: 25,
condition: "count > 100"
})
// 3. Resume until condition met
resume_execution({ includeScopes: true, includeConsole: true })
// 4. Inspect when paused
inspect_scopes({ maxProps: 15 })
evaluate_expression({ expr: "count" })
// 5. Continue
resume_execution()
// Resume with console capture
const result = await resume_execution({ includeConsole: true });
// Or read explicitly
const consoleOutput = await read_console({ format: "text" });
// Get full call stack
list_call_stack({ depth: 20, includeThis: true })
// Inspect different frames
inspect_scopes({ frameIndex: 0 }) // Current frame
inspect_scopes({ frameIndex: 1 }) // Caller frame
inspect_scopes({ frameIndex: 2 }) // Caller's caller
// Evaluate in different frame context
evaluate_expression({ expr: "localVar", frameIndex: 1 })
// Pause on all exceptions
set_exception_breakpoints({ state: "all" })
// Resume and wait for exception
const result = await resume_execution({ includeStack: true })
// When paused on exception:
get_pause_info() // Shows exception details
list_call_stack({ depth: 10 })
inspect_scopes({ maxProps: 20 })
// 1. Start debugging an Express server
{
"tool": "start_node_debug",
"params": {
"scriptPath": "/path/to/server.js",
"env": {
"PORT": "3000",
"NODE_ENV": "development"
}
}
}
// 2. Set breakpoint in route handler
{
"tool": "set_breakpoint",
"params": {
"filePath": "/path/to/routes/users.js",
"line": 15
}
}
// 3. Set logpoint to track requests
{
"tool": "add_logpoint",
"params": {
"filePath": "/path/to/middleware/auth.js",
"line": 8,
"message": "Auth check for user: {req.user.id}"
}
}
// 4. Pause only on errors
{
"tool": "set_exception_breakpoints",
"params": { "state": "uncaught" }
}
// 5. Resume and make HTTP request (externally)
{
"tool": "resume_execution",
"params": {
"includeScopes": true,
"includeConsole": true
}
}
// 6. When paused at breakpoint, inspect request
{
"tool": "evaluate_expression",
"params": {
"expr": "req.body",
"returnByValue": true
}
}
{
"tool": "evaluate_expression",
"params": {
"expr": "req.headers['authorization']"
}
}
// 7. Check database query in scope
{
"tool": "inspect_scopes",
"params": { "maxProps": 30 }
}
// 8. Step into database function
{
"tool": "step_into",
"params": { "includeStack": true }
}
// 9. Continue execution
{
"tool": "resume_execution"
}
// 10. Stop when done
{
"tool": "stop_debug_session"
}
// app.js
async function fetchUserData(userId) {
const user = await db.findUser(userId);
const posts = await db.findPosts(user.id);
return { user, posts };
}
// Debugging session:
// 1. Start
start_node_debug({ scriptPath: "/path/to/app.js" })
// 2. Break at async function
set_breakpoint({ filePath: "/path/to/app.js", line: 2 })
// 3. Resume to breakpoint
resume_execution({ includeScopes: true })
// 4. Check userId parameter
evaluate_expression({ expr: "userId" })
// 5. Step over await (resumes until promise resolves)
step_over({ includeScopes: true, includeConsole: true })
// 6. Inspect resolved user object
evaluate_expression({ expr: "user" })
get_object_properties({ objectId: "object:user123", maxProps: 20 })
// 7. Continue
resume_execution()
filePath parametersfile:// URLs for CDP// ✅ Correct
set_breakpoint({ filePath: "/home/user/project/src/app.js", line: 42 })
// ❌ Wrong (relative path)
set_breakpoint({ filePath: "./src/app.js", line: 42 })
Problem: start_node_debug fails or hangs
Solutions:
scriptPath is an absolute pathconsole.log('test'))// Test with minimal script
start_node_debug({ scriptPath: "/tmp/test.js" })
// test.js content: console.log('Hello');
Problem: Breakpoint set but execution doesn't pause
Solutions:
list_scripts to confirm)// List all scripts to find exact URL
list_scripts()
// Use URL regex instead of file path
set_breakpoint_condition({
urlRegex: ".*app\\.js$",
line: 42
})
Problem: Breakpoints in TypeScript sources don't work
Solutions:
"sourceMap": true in tsconfig.json).map files exist alongside compiled JS// If TypeScript breakpoints fail, use compiled JS path
set_breakpoint({ filePath: "/path/to/dist/app.js", line: 58 })
Problem: Objects truncated or not showing all properties
Solutions:
maxProps parameterget_object_properties to drill downevaluate_expression// Get more properties
inspect_scopes({ maxProps: 100 })
// Or drill into specific object
evaluate_expression({ expr: "largeObject.specificProperty" })
Problem: Console logs not captured
Solutions:
includeConsole: true on step/resume operationsread_console explicitly after stepping// Include console in step
step_over({ includeConsole: true })
// Or read explicitly
read_console({ format: "text" })
Problem: Zombie Node.js processes after debugging
Solutions:
stop_debug_session when doneps aux | grep node# Kill orphaned debugger processes
pkill -f "node --inspect-brk"
Problem: Execution pauses unexpectedly
Solutions:
set_exception_breakpoints)get_pause_info to understand why paused// Check why paused
get_pause_info({ format: "text" })
// Disable exception breaks if too noisy
set_exception_breakpoints({ state: "none" })
includeScopes, includeStack, includeConsole on steps for richer debuggingblackbox_scriptsstate: 'uncaught' to avoid pausing on handled exceptionsstop_debug_session when donepauseId for context-specific operationsframeIndex to evaluate expressions in specific call framesThe MCP server itself doesn't require environment variables, but your Node.js scripts may:
{
"tool": "start_node_debug",
"params": {
"scriptPath": "/path/to/app.js",
"env": {
"DATABASE_URL": process.env.DATABASE_URL,
"API_KEY": process.env.API_KEY,
"NODE_ENV": "development"
}
}
}
Never hardcode secrets in debugging params — reference environment variables or use a .env file in your project.
MIT
Agent 工作流
创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.
Agent 工作流
使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.
Agent 工作流
管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.
Agent 工作流
Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…