Chrome DevTools for Agents 1.0 正式发布 https://developer.chrome.com/blog/devtools-for-agents-v1…

它在真实浏览器中观察行为、检查输出，让 Agent “能看见浏览器”，有三种接入方式：

1. MCP server：将 LLM 连接到 DevTools 调试能力的标准协议
2. CLI：Token 更省的替代方案，支持 Agent 把动作打包成脚本批量执行
3. Agent skills：教 Agent 何时、如何调用具体工具的专家指令 (如无障碍、性能调试)

共开放了七个能力

1. 自动化质量审计：Agent 可直接跑 Lighthouse，覆盖无障碍、SEO、最佳实践、agentic browsing。可作为“质量门“，拦截阻塞性问题进入生产。
2. 真实用户环境模拟：窗口尺寸、地理位置、网络/CPU 节流均可由 Agent 操控，无需手动调整浏览器即可测试响应式与移动端行为 (如汉堡菜单)。
3. Chrome 扩展开发与调试：安装、reload、触发扩展动作，介入 background script 与扩展页面，自动化“保存-刷新“循环。
4. WebMCP 工具调试：配合 WebMCP Origin Trial。让站点向 Agent 暴露结构化工具，Agent 不再靠 DOM 猜测意图，而是直接列出、调用、验证工具，显著降低集成门槛。
5. 内存泄漏检测：支持堆快照，识别 detached DOM 节点等典型泄漏。配合内存调试 skill，Agent 扮演性能专家角色。
6. Auto-connect 会话接管：可把当前已登录的浏览器上下文交给 Agent，而非让其打开沙箱实例。适合调试需要鉴权的页面 (如后台仪表板)，省去重新登录。
7. 第三方开发者工具暴露内部状态：Web 应用可主动向 Agent 暴露内部状态与组件细节，使调试建议基于真实运行时数据，而非黑盒推断。

Streamline your AI coding workflow with Chrome DevTools for agents 1.0 | Blog | Chrome for Developers

Source: https://developer.chrome.com/blog/devtools-for-agents-v1

Published: May 19, 2026

AI coding tools are incredibly powerful at writing code, but they are often disconnected from its execution. They can generate a complex web app, but they cannot observe its behavior or inspect its output in a live browser.

Chrome DevTools for agentsprovides your coding agent with the visibility it needs to verify, debug, and optimize code in real time. A couple months agowe showed it for the first time, and now we are excited to announce that Chrome DevTools for agents is now available as a stable 1.0 release.

Connect your agent to the browser

Chrome DevTools for agents allows your agent to experience your site like a real user would. This stable release includes several ways for you and your agent to interface with Chrome:

• Model Context Protocol (MCP) server: A server that connects Large Language Models (LLMs) to DevTools debugging capabilities.
• Command-line interface (CLI): A token-efficient alternative that allows agents to batch actions into scripts.
• Agent skills: Expert instructions that teach your agent how and when to use specific tools for tasks like accessibility or performance debugging.

Debug, emulate, and audit automatically

With the 1.0 release, your coding agent can now perform advanced debugging tasks that were previously manual.

Automate quality audits

Your agent can nowrun Lighthouse auditsto evaluate website quality. It can identify issues with accessibility, SEO, best practices, and agentic browsing. It’s like having a linter that understands the runtime. Used as a quality gate, it turns your coding agent into an expert that catches critical blockers before they reach production.

Emulate real-world user experiences

Your agent can use DevTools for agents to test how a site looks and performs on various devices or in different locationsusing emulation tools. It can resize windows, simulate geolocations, and throttle network and CPU speeds to mimic real-world conditions. This way, your agent can test mobile-specific behaviors, like burger menus, without you having to manually resize your browser.

Develop and debug Chrome Extensions

You can also let your agent not only help you develop Chrome Extensions but also debug them. Your agent can install, reload, and trigger extension actions directly. This helps automate the frequent “save-and-refresh” cycle during development. It can also dive into background scripts and extension pages to help you fix bugs in more complex browser setups.

Debug and test WebMCP tools

We’re also bringing deep visibility to theWebMCP(Web Model Context Protocol) Origin Trial to help you implement this new API.

Instead of having your agent rely on signals to infer a plan for navigating the DOM, it can directly interact with structured tools that you expose using WebMCP. This makes it easier to develop, test and debug these tools: your agent can list, programmatically invoke, and validate their correctness in real time. This streamlines the process of adding WebMCP support to your site and testing it.

Detect and debug memory leaks

We have also added dedicated tools for memory analysis. Your agent can now take heap snapshots to identify memory leaks, such as detached DOM nodes. By using specialized memory leak debugging skills, the agent acts as a performance expert to help you keep your application lean and fast.

Hand over sessions with auto-connect

You canshare your current browser context with an agent, instead of the agent opening its own sandboxed browser instance (which is how DevTools for agents usually gives agents access to the browser). This is perfect for debugging things that require a login, like an authenticated dashboard, where you want the AI agent to take over the technical investigation without you having to re-authenticate everything.

Expose internal state with third-party developer tools

Third-party developer tools allow your web application to share internal state and component details directly with AI agents. By providing access to data that is usually hidden from standard analysis, these tools give agents the context they need to understand complex logic. This results in more accurate debugging suggestions based on how the application is actually behaving in the browser.

Get started

You can install the stable release usingnpmor configure it directly in your agent of choice:

Antigravity

Chrome DevTools for agents comes pre-bundled withAntigravity 2.0. You can start using it immediately with thebrowser sub-agent. Try using a slash command such as:

/browser Navigate to the Google homepage

To access specialized agent skills, we recommend installing the DevTools plugin during theBuild with Googlestep of the initial setup or within the application settings. For more information, check out theAntigravity Browser Subagent documentation.

Gemini CLI

To install the MCP package and skills as an extension, use the following command:

$ gemini extensions install --auto-update https://github.com/ChromeDevTools/chrome-devtools-mcp

Claude Code

To install Chrome DevTools for agents as a Claude Code Plugin, use the following slash commands in Claude Code. Add the marketplace registry:

/plugin marketplace add ChromeDevTools/chrome-devtools-mcp

And install the plugin from the marketplace registry:

/plugin install chrome-devtools-mcp@chrome-devtools-plugins

For more detailed guides and example prompts, visit ourChrome DevTools for agents documentation. You can also explore thesource code on GitHub.