BrowserStack MCP Server

The BrowserStack MCP Server is the official integration that connects BrowserStack's cloud testing infrastructure to AI clients like Claude Desktop, Cursor, VS Code (GitHub Copilot), and Cline. Through natural language prompts, agents can spin up live manual sessions on real iOS, Android, and desktop devices, execute Selenium and Playwright suites via the BrowserStack SDK, pull failure logs, and inspect automation screenshots, all without leaving the editor.

In addition to running tests, the server exposes BrowserStack's test management product (projects, folders, test cases, test runs, and results) and a set of AI agents for accessibility scanning, self-healing selectors, low-code automation step generation, and producing test cases from product requirement documents. This makes it useful both for QA engineers managing test suites and for developers debugging or writing tests directly from their IDE.

The server is open source under AGPL-3.0, published as the npm package @browserstack/mcp-server, and authenticates with a standard BrowserStack username and access key. A hosted remote MCP endpoint with OAuth is also available at https://mcp.browserstack.com/mcp for environments that cannot run a local process.

Tools

Prerequisites

• Node.js 18.0 or higher (v22.15.0 LTS recommended)
• An active BrowserStack account
• Your BROWSERSTACK_USERNAME and BROWSERSTACK_ACCESS_KEY from Account Settings

Install

The server is distributed via npm as @browserstack/mcp-server and is launched on demand with npx. No global install is needed.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or the equivalent on Windows:

{
  "mcpServers": {
    "browserstack": {
      "command": "npx",
      "args": ["-y", "@browserstack/mcp-server@latest"],
      "env": {
        "BROWSERSTACK_USERNAME": "<username>",
        "BROWSERSTACK_ACCESS_KEY": "<access_key>"
      }
    }
  }
}

Cursor

Add to .cursor/mcp.json:

{
  "mcpServers": {
    "browserstack": {
      "command": "npx",
      "args": ["-y", "@browserstack/mcp-server@latest"],
      "env": {
        "BROWSERSTACK_USERNAME": "<username>",
        "BROWSERSTACK_ACCESS_KEY": "<access_key>"
      }
    }
  }
}

VS Code (GitHub Copilot)

Add to .vscode/mcp.json:

{
  "servers": {
    "browserstack": {
      "command": "npx",
      "args": ["-y", "@browserstack/mcp-server@latest"],
      "env": {
        "BROWSERSTACK_USERNAME": "<username>",
        "BROWSERSTACK_ACCESS_KEY": "<access_key>"
      }
    }
  }
}

Remote MCP (no local install)

BrowserStack hosts an OAuth-authenticated remote endpoint at https://mcp.browserstack.com/mcp. Use it in clients that support HTTP MCP servers when you cannot run a local Node process. Note that local/VPN-only test targets are not reachable via the remote server.

• Kick off a manual App Live or Live session on a specific real device or browser from inside your IDE while reviewing a PR.
• Pull failure logs and automation screenshots for a failing Selenium or Playwright build and have the agent suggest a fix in the source code.
• Run an accessibility scan against a staging URL and get a prioritized list of WCAG violations to remediate.
• Generate manual test cases from a PRD, convert them to low-code automation steps, then organize them into projects, folders, and runs.
• Reduce flakiness in existing automation by fetching AI self-healed selectors for broken locators.

• "Open a Live session on iPhone 15 Pro running Safari and load staging.example.com."
• "Get the failure logs for my last BrowserStack Automate build and tell me what likely broke."
• "Run an accessibility scan on https://example.com and summarize the critical WCAG issues."
• "Create a test run in the Checkout project from all test cases tagged 'smoke' and assign it to me."
• "Upload this PRD and generate manual test cases for the new signup flow."

• Official, actively maintained server from BrowserStack with frequent releases.
• Broad coverage across Live, App Live, Automate, App Automate, Test Management, Accessibility, and AI agent products.
• Works with all major MCP clients (Claude Desktop, Cursor, VS Code Copilot, Cline) and offers a hosted OAuth remote endpoint.
• Open source under AGPL-3.0, so behavior and tool definitions are inspectable.

• Requires a paid BrowserStack account to actually run tests on real devices and browsers.
• Behavior is non-deterministic in places because tool calls are LLM-driven; the README explicitly flags this.
• The hosted remote MCP cannot reach local or VPN-only test targets, forcing local install for those flows.

• Playwright MCP for local browser automation without a cloud device farm.
• Puppeteer MCP server for scripted Chromium control from agents.
• Sauce Labs and LambdaTest as alternative cross-browser cloud platforms (no first-party MCP server at parity with BrowserStack's at time of writing).