BROWSER INFRASTRUCTURE · FOR AI AGENTS
BrowserStack MCP
Run agent-directed tests on hosted real mobile devices and desktop browsers.
BrowserStack connects agents to its testing infrastructure through MCP. Agents can start a device or browser session, interact with an app or website, capture screenshots and inspect session results without maintaining their own device lab.
Your integration checklist
- Capabilities
- What you need from your owner
- Create a BrowserStack account and obtain the product licence for the workflow: App Automate for mobile device control or Automate for desktop browser control. Provide an authorized app build or test URL and connect the MCP client.
- Authentication
- The remote MCP endpoint https://mcp.browserstack.com/mcp uses OAuth. A local server is also available and uses BrowserStack credentials; local app-file upload requires MCP_UPLOAD_BASE_DIR to identify the permitted upload folder.
- Your first task
- Run one known sign-in or navigation flow on a specified device and OS, capture the result, then end the session with an explicit pass or fail reason.
- Billing & payment
- Access depends on the underlying BrowserStack product and plan, including concurrency and device capacity. Trial availability does not establish unrestricted real-device usage. Check App Automate and Automate entitlements separately.
- Agent payment support
- Not verified. Programmatic usage does not establish autonomous checkout support.
- Operational limits
- Device availability and parallel capacity depend on the account. Sessions have idle and total-duration limits; videos are complete after the session ends. A successful action sequence does not establish broad device coverage or accessibility compliance.
ORIGINAL AI AGENT STORE INTEGRATION PLAYBOOK
Pin the build and device before judging a test result
Editorial recommendation · Reviewed 2026-11-10
When to choose this service
Choose BrowserStack when an agent needs evidence from actual device and browser environments beyond its local runtime. Local browser testing may suffice for a narrow desktop-only change.
Physical mobile devices, maintained operating-system environments and remote session infrastructure are scarce resources. Writing a test script cannot reproduce the behavior of hardware the agent does not possess.
Suggested workflow
- Record the exact app build or website revision and select the required device, OS and browser. Confirm the account can start that session.
- Run a small flow with expected outcomes and collect screenshots or logs at decisive steps. Separate application failures from unavailable devices and setup errors.
- Mark the outcome with a concrete reason, end the session and retain the completed evidence. Expand the device matrix only where it answers a compatibility question.
Acceptance checks before real work
- An app build identifier must match the code under review.
- A session timing out is not automatically a product defect.
- A video link may be incomplete until the session has ended.
If the workflow fails
If a session cannot start, check product entitlement, concurrency and device availability first. For a failed test, reproduce the smallest failing step on the same build before changing the application.