Browserstack

io.github.browserstack由 GitHub 账号 browserstack 发布v2.6.11更新于 Nov 19, 2026

BrowserStack's Official MCP Server

已验证Streamable HTTP可网页运行浏览器自动化开发者工具安全与监控

概览

AI 生成的概览

BrowserStack 官方 MCP 服务器,让助手用自然语言运行、调试和管理 BrowserStack 测试、测试用例与负载测试。

可以做什么
  • 通过五个核心工具发现并执行 BrowserStack 工作流:列出产品、说明实体与能力、搜索能力并调用
  • 在 BrowserStack 上运行和调试自动化测试(Playwright、Selenium、Jest),获取日志与根因分析,并把测试套件迁移到其基础设施
  • 在真实浏览器和设备上打开网站与移动应用,包括托管在 localhost 的站点
  • 扫描页面的无障碍问题并获取修复建议
  • 创建和管理测试用例、测试计划与测试运行,并更新测试结果
  • 运行和分析负载测试、比较多次运行、查看 VU 小时配额,并使用 BrowserStack AI 代理生成测试、低代码编写和自动修复
适用场景
适合已使用 BrowserStack、希望让 IDE 或聊天中的助手直接运行和调试测试套件、在真实设备上复现缺陷、管理测试用例,或执行负载与无障碍测试而不必切换工具的场景。
运行要求
需要 BrowserStack 账号及用户名和访问密钥,托管远程服务器可用 OAuth 登录。远程端点无需本地安装;本地 npm 包需要 Node.js 22.23.5 或更高版本。文件与应用上传类工具需要设置 MCP_UPLOAD_BASE_DIR 环境变量,指向允许上传的目录。
安装前请注意
服务器持有 BrowserStack 凭据(BROWSERSTACK_USERNAME 与 BROWSERSTACK_ACCESS_KEY)或 OAuth 会话,应按机密对待。修改数据的操作需要你批准,但助手仍可创建项目、测试用例和测试运行,并启动或停止会消耗付费 VU 小时的负载测试。远程服务器无法访问 localhost 或仅限 VPN 的应用,且工具调用依赖大模型,结果可能不确定。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Browserstack,将其添加到工作区。
  2. 连接 所需的 凭据。它们会被私密存储,绝不会在公开页面上展示。
  3. 为需要使用其工具的对话启用该服务。

Web executable,通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

交给智能体安装

把这段提示词粘贴到 Claude Code、Codex、Cursor 等智能体,或 SourceWeft 对话中。智能体会读取该服务器的安装指南,先向你展示发布者、端点和登录方式,经你同意后再把它添加到你的客户端。密钥不会写进任何命令。

阅读 https://sourceweft.com/mcp/io.github.browserstack%2Fmcp-server/install.md,并按其中说明把这个 MCP 服务器接入我的客户端。修改任何配置之前,先告诉我它的发布者、端点和登录方式,等我确认。不要把 API 密钥写进命令或我会提交的文件;问我密钥应从哪里获取。

查看智能体所依据的安装指南

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。 客户端会在首次连接时完成 OAuth 登录。

{
  "mcpServers": {
    "browserstack": {
      "type": "http",
      "url": "https://mcp.browserstack.com/mcp"
    }
  }
}

README

BrowserStack MCP Server

Comprehensive Test Platform

Manage test cases, execute manual or automated tests, debug issues, and even fix code—directly within tools like Cursor, Claude, or any MCP-enabled client, using plain English.

Test from anywhere:

Easily connect the BrowserStack Test Platform to your favourite AI tools, such as IDEs, LLMs, or agentic workflows.

Test with natural language:

Manage, execute, debug tests, and even fix code using plain English prompts.

Reduced context switching:

Stay in flow—keep all project context in one place and trigger actions directly from your IDE or LLM.

⚡️ One Click MCP Setup

Click on the buttons below to install MCP in your respective IDE or assistant:

[Install in VS Code]  [Add to Copilot]  [Add to Cursor]  [Add to Claude]  [Add to ChatGPT]

Note: The hosted Remote MCP server (mcp.browserstack.com) is now fully stateless over Streamable HTTP — any Streamable-HTTP client (Claude, Cursor, VS Code, ChatGPT) connects with no configuration change.

Kiro: a Kiro power for this server lives in integrations/kiro-power. In the Kiro IDE, open the Powers panel → Add Custom Power → Import power from GitHub and enter https://github.com/browserstack/mcp-server/tree/main/integrations/kiro-power, or clone this repository and use Import power from a folder with that directory.

Prerequisites

For manual setup in any MCP client, see Install the MCP server locally or Set up the remote MCP server.

Note : Ensure you are using Node version >= 22.23.5
  • Check your node version using node --version. Recommended version: v24.22.3 (current), minimum v22.23.5.
  • To Upgrade Node :
    1. On macOS (Homebrew) - brew update && brew upgrade node or if using (nvm) - nvm install 24.22.3 && nvm use 24.22.3 && nvm alias default 24.22.3
    1. On Windows (nvm-windows) : nvm install 24.22.3 && nvm use 24.22.3
  • 👉 Or directly download the Node.js LTS Installer

🪄 Agent-assisted setup

Most AI agents can add the BrowserStack MCP server, connect to it, and walk you through sign-in without manual config. Paste this prompt into your agent:

text
Add the BrowserStack MCP server to this agent, following the setup guide athttps://www.browserstack.com/docs/browserstack-mcp-server/get-started/remote-mcp-serverand using the server URL https://mcp.browserstack.com/mcp. Once it is added, start theBrowserStack sign-in flow so I can log in.

To add it yourself, use your client's command or menu:

ClientCommand or configuration
Claude Codeclaude mcp add --scope user --transport http browserstack https://mcp.browserstack.com/mcp, then run /mcp in a session and sign in
Claude Desktop / claude.aiSettings → Connectors → Add custom connector, with the URL https://mcp.browserstack.com/mcp
ChatGPTInstall the BrowserStack plugin
Codexcodex mcp add browserstack --url https://mcp.browserstack.com/mcp, then codex mcp login browserstack to sign in
VS Code / GitHub Copilotcode --add-mcp '{"name":"browserstack","type":"http","url":"https://mcp.browserstack.com/mcp"}', then start the server from the MCP view and sign in
CursorAdd "browserstack": { "url": "https://mcp.browserstack.com/mcp" } under mcpServers in ~/.cursor/mcp.json, then sign in from Cursor Settings → MCP
Any other MCP clientPoint it at https://mcp.browserstack.com/mcp

💡 Usage Examples

📱 Manual App Testing

Test mobile apps on real devices across the latest OS versions. Reproduce bugs and debug crashes without setup hassles. Below are some sample prompts to use your mobile apps on BrowserStack's extensive cloud of real devices

bash
# Open app on specific device"open my app on a iPhone 15 Pro Max"
# Debug app crashes"My app crashed on Android 14 device, can you help me debug?"
  • Unlike emulators, test your app's real-world performance on actual devices. With advanced App-Profiling features, you can debug crashes and performance issues in real-time.
  • Access all major devices and OS versions from our device grid, We have strict SLAs to provision our global datacenters with newly released devices on launch day.

🌐 Manual Web Testing

Similar to the app testing, you can use the following prompts to test your websites on BrowserStack's extensive cloud of real browsers and devices. Don't have Edge browser installed on your machine ? We've got you covered!

bash
# Test your websites"open my website hosted on localhost:3001 on Edge""open browserstack.com on latest version of Chrome"
  • Test websites across different browsers and devices. We support every major browser across every major OS.
  • Seamlessly test websites hosted locally on your machine, no need to deploy to a remote server!

🧪 Automated Testing (Playwright, Selenium, A11y and more..)

Auto-analyze, diagnose, and even fix broken test scripts right in your IDE or LLM. Instantly fetch logs, identify root causes, and apply context-aware fixes. No more debugging loops. Below are few example prompts to run/debug/fix your automated tests on BrowserStack's Test Platform.

Note: When fetching Root Cause Analysis (RCA) for a test, the server returns the suggested fix as a proposal only. It never applies code changes automatically — your assistant must present the suggestion and wait for your explicit approval before editing any files.

bash
#Port test suite to BrowserStack"Setup test suite to run on BrowserStack infra"
#Run tests on BrowserStack“Run my tests on BrowserStack”
#AI powered debugging of test failures"My App Automate tests have failed, can you help me fix the new failures?"
  • Fix test failures reported by your CI/CD pipeline by utilising our industry leading Test Observability features. Find more info here.
  • Run tests written in Jest, Playwright, Selenium, and more on BrowserStack's Test Platform

🌐 Accessibility

Catch accessibility issues early with automated, local a11y scans. Get one-click, AI-suggested fixes. No docs hunting, no CI surprises. Ensure WCAG and ADA compliance with our Accessibility Testing tool

bash
#Scan accessibility issues while development"Scan & help fix accessibility issues for my website running locally on localhost:3000"
#Scan accessibility issues on production site“Run accessibility scan & identify issues on my website - www.bstackdemo.com”

📋 Test Management

Create and manage test cases, create test plans and trigger test runs using natural language. Below are a few example prompts to utilise capabilities of BrowserStack's Test Management with MCP server.

bash
# Create project & folder structure"create new Test management project named My Demo Project with two sub folders - Login & Checkout"
# Add test cases"add invalid login test case in Test Management project named My Demo Project"
# List added test cases "list high priority Login test cases from Test Management project - My Demo Project"
# Create test run"create a test run for Login tests from Test Management project - My Demo Project"
# Update test results"update test results as passed for Login tests test run from My Demo Project"

⚡ Load Testing

Run and analyse load tests at scale without leaving your IDE or chat — find tests, start and stop runs, read reports and AI Insights, compare runs, and check your VU-hours. Works with API (JMeter, k6, Gatling, Locust), browser (Playwright, Selenium, WebdriverIO, Nightwatch) and hybrid load tests on BrowserStack's Load Testing. These go through the core tools.

bash
# Find a test"find my load test called checkout-api"
# Run it"run checkout-api for 5 minutes with 100 virtual users"
# Check on it"is my checkout-api run finished yet?"
# Read the results"what was the p95 response time and error rate of the last checkout-api run?"
# Compare two runs of the same test"did the latest checkout-api run regress compared with the previous one?"
# Check quota before a big run"how many VU-hours do I have left, and what would a 30-minute run with 500 VUs cost?"

🧪 Access BrowserStack AI agents

Generate test cases from PRDs, convert manual tests to low-code automation, and auto-heal flaky scripts powered by BrowserStack’s AI agents, seamlessly integrated into your workflow. Below are few example prompts to access Browserstack AI agents

bash
#Test case generator agent"With Browserstack AI, create relevant test cases for my PRD located at /usr/file/location"

#Low code authoring agent“With Browserstack AI, automate my manual test case X, added in Test Management”

#Self healing agent“Help fix flaky tests in my test script with Browserstack AI self healing”

🧭 Core Tools

Five core tools let your AI assistant find and run any BrowserStack workflow. Your assistant loads every tool an MCP server exposes at the start of each session, so one tool per workflow would fill your context window and slow down the whole session. Instead, your assistant uses the core tools to search a catalog of BrowserStack tools in plain language and run the one it needs.

ToolDescription
listProductsLists the BrowserStack products in the catalog, with a short summary of what each product covers
describeEntityExplains one entity: its aliases, ID format, parent and related entities, and the capabilities that act on it
searchCapabilityFinds the capabilities that match a plain-language request, with everything your assistant needs to run each one
describeCapabilityReturns the full details of one capability, such as its parameters, request body and response format
invokeCapabilityRuns one capability and returns the product's response unchanged. Changes to data need your approval

See Core tools for parameters and how changes to data are approved.

All tools — See docs/TOOLS.md for every tool organised by product, or BrowserStack MCP tools & workflows for product workflows.

File & app uploads: tools that upload a local file/app (uploadProductRequirementFile, takeAppScreenshot, runAppTestsOnBrowserStack, runAppLiveSession) require the MCP_UPLOAD_BASE_DIR env var set to a directory containing those files; uploads are restricted to it.

🚀 Remote MCP Server

Remote MCP comes with all the functionalities of an MCP server without the hassles of complex setup or local installation.

Key benefits:

  • ✅ Works seamlessly in enterprise networks without worrying about firewalls or binaries or where local installation is not allowed.

  • ✅ Secure OAuth integration – no password sharing or manual credential handling.

Limitations:

  • ❌ No Local Testing support (cannot test apps behind VPNs, firewalls, or localhost). If you have to do Local Testing, you would have to use a BrowserStack Local MCP server.
  • ❌ Latency can be slightly higher, but nothing considerable — you generally won’t notice it in normal use.

⚠️ Important Notes

  • The BrowserStack MCP Server is under active development and currently supports a subset of the MCP spec. More features will be added soon.
  • Tool invocations rely on the MCP Client which in turn relies on an LLM, hence there can be some non-deterministic behaviour that can lead to unexpected results. If you have any suggestions or feedback, please open an issue to discuss.

📝 Contributing

We welcome contributions! Please open an issue to discuss any changes you'd like to make. 👉 Click here to view our Contributing Guidelines

📞 Support

For support, please:

🚀 More Features Coming Soon

Stay tuned for exciting updates! Have any suggestions? Please open an issue to discuss.

来源:README.md,提交 577c98f

工具

0
该服务器需要登录,因此 SourceWeft 无法在没有账号的情况下读取其工具,也未收录。已连接账号所能看到的内容可能因账号及其权限而异。

版本历史

24
  1. v2.6.11最新Nov 19, 2026
  2. v2.6.10Nov 19, 2026
  3. v2.6.9Nov 16, 2026
  4. v2.6.8Nov 12, 2026
  5. v2.6.7Nov 12, 2026
  6. v2.6.6Nov 11, 2026
  7. v2.6.5Nov 7, 2026
  8. v2.6.1Nov 4, 2026
  9. v2.6.0Nov 4, 2026
  10. v2.5.0Nov 3, 2026
  11. v2.4.5Oct 30, 2026
  12. v2.4.1Oct 29, 2026
  13. v2.2.8Oct 26, 2026
  14. v2.2.7Oct 23, 2026
  15. v2.2.6Oct 23, 2026
  16. v2.2.5Oct 22, 2026
  17. v2.2.4Oct 19, 2026
  18. v2.2.3Oct 16, 2026
  19. v2.2.2Oct 14, 2026
  20. v2.1.1Oct 5, 2026
  21. v2.1.0Sep 30, 2026
  22. v2.0.0Sep 29, 2026
  23. v1.5.1Sep 28, 2026
  24. v1.0.0Sep 16, 2026

评分

已满足适用检查项的 85%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通必需25 分
  • 已声明认证方式必需10 分
  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分