AI Agent 云端浏览器基础设施
为 Agent 提供可弹性伸缩的云端浏览器
Lexmount 把浏览器从本地软件变成可调用、可管理、可观测的云端资源。Agent 可以按需创建、连接、观测和回放浏览器任务,让网页任务突破本地硬件限制,获得更强的执行能力。
Browser CLI Agent 提示词
选择你正在使用的 Agent,再复制对应的 Lexmount Browser 安装和使用提示词。
Use Lexmount Browser through browser-cli for remote browser work.
Goal:
- Prefer browser-cli commands over handwritten Playwright.
- If Lexmount credentials are missing, guide me through browser-cli auth login.
- Never ask me to paste API keys or full direct browser URLs into chat.
- API keys must not appear in localhost callback URLs; browser-cli exchanges a one-time code with PKCE.
Official site:
- Lexmount Browser console: https://browser.lexmount.cn
- Use it to sign in, register, authorize CLI access, and manage browser projects.
Install:
1. Check whether browser-cli is already installed: browser-cli --version
2. If browser-cli is missing and uv is missing, install uv:
curl -LsSf https://astral.sh/uv/install.sh | sh
3. Install or update browser-cli to 0.3.8 or newer:
uv tool install --force git+https://github.com/lexmount/browser-cli.git
4. Install or update the packaged Codex skill:
browser-cli skill status
browser-cli skill install --force
browser-cli skill status
5. Verify installation:
browser-cli --version
browser-cli version
Connect:
1. Run: browser-cli auth status
2. If credentials are missing or unusable, run:
browser-cli auth login --open --connect-base-url https://browser.lexmount.cn --client-name Codex
3. Complete the https://browser.lexmount.cn Connect from Agent flow in the opened browser.
4. Confirm auth login reports authenticated=true or credentials_saved=true.
5. After login succeeds, run: browser-cli doctor --json
Installation result:
- browser-cli is installed or updated.
- The Lexmount Browser Codex skill is installed under the Codex skills directory.
- The API credential is saved by browser-cli auth login; do not paste API keys into chat.
Cleanup:
- To remove saved local browser-cli credentials, run:
browser-cli auth clear-credentials
- If that command returns unset_env_commands, run those commands in the parent shell too.
- Use browser-cli auth logout --revoke only when you intentionally want remote token revocation.
Use:
- Create browser sessions with browser-cli session create.
- Inspect first with browser-cli action observe or browser-cli action interactive-snapshot.
- Prefer semantic actions such as click-role, click-text, fill-label, wait-role, get-text-role, exists-role, hover-role, press-role, and screenshots.
- If a page requires login, captcha, SMS verification, or QR-code scanning, do not just say that it is blocked. Get the current session inspect_url / inspectUrl from the session metadata or latest command output, open that URL with the available local/embedded browser, and ask the user to complete the handoff there. If no inspect URL is available, report that blocker explicitly.
- Do not paste full direct browser connect URLs into chat. Use inspect_url / inspectUrl for human handoff; use --reveal-connect-url only in a trusted local terminal when explicitly needed for debugging.
- For repeatable tasks, inspect browser-cli commands --workflow case_file_task and use case scaffold/run.
- Close temporary sessions unless I explicitly ask to keep them open.WebFetch Agent 提示词
选择你正在使用的 Agent,再复制对应的 WebFetch 安装和使用提示词。
Use Lexmount WebFetch through webfetch-cli when the task only needs page extraction or DOM dump.
Goal:
- Prefer webfetch-cli for lightweight page extraction instead of opening a remote browser session.
- Use browser-cli only when interactive browser state, clicking, screenshots, or live session control is needed.
- Never ask me to paste API keys into chat.
- API keys must not appear in localhost callback URLs; webfetch-cli exchanges a one-time code with PKCE.
Official site:
- Lexmount Browser console: https://browser.lexmount.cn
- Use it to sign in, register, authorize CLI access, and manage WebFetch project credentials.
Install:
1. Check whether webfetch-cli is already installed: webfetch-cli --version
2. If webfetch-cli is missing and uv is missing, install uv:
curl -LsSf https://astral.sh/uv/install.sh | sh
3. Install or update webfetch-cli to 0.1.5 or newer:
uv tool install --force git+https://github.com/lexmount/webfetch-cli.git
4. Install or update the packaged Codex skill:
webfetch-cli skill status
webfetch-cli skill install --force
webfetch-cli skill status
5. Verify installation:
webfetch-cli --version
webfetch-cli capabilities --json
Connect:
1. Run: webfetch-cli auth status
2. If credentials are missing or unusable, run:
webfetch-cli auth login --open --connect-base-url https://browser.lexmount.cn --client-name Codex
3. Complete the https://browser.lexmount.cn Connect from Agent flow in the opened browser.
4. Confirm auth login reports authenticated=true or credentials_saved=true.
5. After login succeeds, run: webfetch-cli doctor --json
Installation result:
- webfetch-cli is installed or updated.
- The Lexmount WebFetch Codex skill is installed under the Codex skills directory.
- The API credential is saved by webfetch-cli auth login; do not paste API keys into chat.
Use:
- Extract structured page content with:
webfetch-cli extract --url https://example.com
- Dump rendered DOM with:
webfetch-cli dump-dom --url https://example.com
- If extract needs an existing DOM snapshot, use:
webfetch-cli extract --dom-id <dom_id>
- Use --format json-full only when debugging full API responses.
Cleanup:
- To remove saved local webfetch-cli credentials, run:
webfetch-cli auth clear-credentialsChatGPT MCP 接入链接
使用与当前站点对应的 Browser MCP 或 WebFetch MCP endpoint。
弹性浏览器池
按需启停
远程 View
人工接管
并行实例
弹性扩展
录像追溯
可回放
Agent Browser Cloud
给 Agent 一组随时可用的云端浏览器
按需创建,并行执行,统一观测。
01
Agent 发起任务
用户用自然语言或自动化代码提出网页操作需求。
02
Lexmount 控制面
统一创建实例、管理凭证,并提供连接入口。
03
云端浏览器池
并行运行多个浏览器,支持观测、接管和录像。
控制面能力
创建、访问、观测和追溯都在同一个平台完成
Lexmount 将浏览器实例生命周期、身份凭证、连接信息、远程 View 和执行录像收束到一个清晰的控制面,方便 Agent 和人工协同使用。
生命周期管理
创建、运行、关闭和回收云端浏览器实例。
凭证与连接入口
统一管理 API 调用身份与浏览器连接信息。
远程 View
打开云端浏览器画面,必要时人工接管操作。
录像追溯
保留任务执行过程,便于排查和复盘问题。



Agent 工作流
多个浏览器并行运行,一个监控画面统一观测
当 Agent 需要同时访问多个网页来源时,Lexmount 可以提供多个云端浏览器并行工作,并把运行画面集中展示,方便观察任务进度。

任务输入
用户用自然语言描述目标。
浏览器执行
Agent 打开云端浏览器并完成网页操作。
实时观测
平台展示多个浏览器的运行画面。
结果与回放
输出结果,并保留录像用于复盘。