BrowserStack MCP Server for Claude

Content

The BrowserStack MCP server connects Claude and other MCP-capable clients to BrowserStack's testing platform. It can start real browser and device sessions, run web and app automation workflows, fetch screenshots and failure logs, launch accessibility scans, and manage BrowserStack Test Management assets from a coding assistant.

This entry is focused on browser and device automation where permission boundaries matter. BrowserStack access is scoped by the account credentials or OAuth identity used by the MCP server, plus the BrowserStack products and projects available to that user. Treat the MCP server as a bridge into a real testing account, not a sandboxed toy browser.

BrowserStack supports both local and remote MCP setup. The local server runs through @browserstack/mcp-server and uses a BrowserStack username and access key. The remote MCP server is hosted by BrowserStack, uses OAuth, and is useful when users cannot or do not want to install a local package.

Features

• Start manual web testing sessions on BrowserStack Live across desktop and mobile browser combinations.
• Start manual app testing sessions on real iOS and Android devices through App Live.
• Set up and run BrowserStack Automate and App Automate workflows for Selenium, Playwright, Appium, Espresso, XCUITest, and related testing stacks.
• Fetch automation screenshots and failure logs for BrowserStack sessions.
• Start web accessibility scans and ask BrowserStack's Accessibility Expert for WCAG-oriented guidance.
• Create projects, folders, test cases, test runs, and test results in BrowserStack Test Management.
• Upload PRDs, screenshots, or test-case files for BrowserStack AI-assisted test generation workflows.
• Use BrowserStack Local through the local MCP setup when testing localhost, staging, VPN, or internal sites is approved.
• Use remote MCP with OAuth when local installation is not practical and Local Testing support is not required.

Use Cases

• Ask Claude to open a website on a specific desktop browser or mobile device combination for manual verification.
• Run existing Playwright, Selenium, Appium, Espresso, or XCUITest suites on BrowserStack infrastructure from an AI-enabled development environment.
• Fetch screenshots or failure logs for a failed Automate or App Automate session and summarize the likely cause.
• Launch an accessibility scan for a target page and turn results into actionable implementation tasks.
• Create or update Test Management projects, cases, runs, and results during a QA workflow.
• Test a local development site on remote browsers through BrowserStack Local when the environment is approved for cloud-device access.

Installation

Claude Code

1. Confirm Node.js is installed: node --version.
2. Create or select a BrowserStack account with access only to the products and projects Claude should use.
3. Add the local MCP server with BrowserStack credentials:

claude mcp add browserstack --env BROWSERSTACK_USERNAME=YOUR_USERNAME --env BROWSERSTACK_ACCESS_KEY=YOUR_ACCESS_KEY -- npx -y @browserstack/mcp-server@latest

1. Restart or refresh the MCP client session.
2. Start with low-impact prompts such as listing available BrowserStack tools or opening a single browser session.

Claude Desktop

1. Open the Claude Desktop MCP configuration file.
2. Add the browserstack server configuration shown below.
3. Replace the username and access key placeholders with BrowserStack account credentials.
4. Restart Claude Desktop and test with a narrow browser or Test Management request.

Configuration

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

Remote MCP

BrowserStack also documents a hosted remote MCP endpoint:

https://mcp.browserstack.com/mcp

Use the remote option when OAuth-based setup is preferred or local package installation is blocked. Use the local MCP server when BrowserStack Local is needed for localhost, staging, VPN, or firewall-restricted test targets.

Examples

Open a browser session

Ask Claude to launch a specific browser and operating system combination for a single approved URL.

Open the staging login page on the latest Chrome version on Windows 11 in BrowserStack Live.

Run automated web tests

Use BrowserStack Automate for an existing browser test suite.

Set up this Playwright test suite to run on BrowserStack and run the checkout smoke tests on Chrome and Firefox.

Fetch failure artifacts

Ask Claude to retrieve screenshots or logs for a known failed session.

Fetch screenshots and failure logs for this BrowserStack Automate session ID and summarize the likely failing step.

Start an accessibility scan

Run a focused scan for one URL instead of broad crawling.

Start an accessibility scan for the pricing page and summarize critical WCAG issues only.

Manage test cases

Create or update test assets only when the target project and folder are specified.

Add a high-priority invalid-login test case to the Checkout QA project under the Login folder.

Security

• Use a dedicated BrowserStack account or access key with only the products, projects, and team permissions needed for MCP workflows.
• Keep credentials in environment configuration or OAuth flows, not prompts or repository files.
• Confirm the target site, device, browser, project, and action before allowing Claude to start sessions, upload app packages, run automation, or update Test Management records.
• Treat BrowserStack Local as an approved network bridge. Do not expose internal apps, VPN-only systems, staging data, or localhost services unless the environment owner has approved cloud testing.
• Avoid unbounded automation loops. Test sessions, screenshots, scans, and retries consume account quota and may create operational noise.

Troubleshooting

Claude cannot start the local server

Confirm Node.js is installed and available to the MCP client process. Then run the package through npx -y @browserstack/mcp-server@latest from the same user environment.

Authentication fails

Verify BROWSERSTACK_USERNAME and BROWSERSTACK_ACCESS_KEY against the BrowserStack Account Settings page. For remote MCP, reconnect through the OAuth flow instead of pasting access keys into prompts.

A tool is unavailable or denied

Check whether the BrowserStack account has access to that product, project, or team feature. The MCP server cannot grant access beyond the connected BrowserStack identity.

Local site testing does not work

Use the local MCP path and configure BrowserStack Local. The remote MCP path does not support Local Testing for localhost, VPN, firewall-restricted, or internal sites.

Session artifacts expose too much data

Restrict the target URL, device session, test suite, and time window. Review screenshots, logs, uploaded files, and generated summaries before sharing them outside the team.