Can AI agents automate BrowserStack?

BrowserStack Test Platform is automatable today through a documented REST API. This record is BrowserStack (Live, Automate, App Automate, accessibility, Test Management, and the first-party MCP server). It is not Playwright, Cypress Cloud, or CircleCI. Automate REST calls use HTTP Basic Auth with a BrowserStack username and access key (curl -u "YOUR_USERNAME:YOUR_ACCESS_KEY"). Opened endpoints include plan, projects, builds, and sessions under https://api.browserstack.com/automate/, returning JSON. That surface manages Automate builds and sessions; it is not a single API for every BrowserStack product. First-party MCP has two transports. Local MCP runs as npx @browserstack/mcp-server with BROWSERSTACK_USERNAME and BROWSERSTACK_ACCESS_KEY. Hosted MCP is at https://mcp.browserstack.com/mcp over HTTP and authenticates with OAuth. Vendor docs and the GitHub README list tools for Test Management, manual web and app testing, Playwright/Selenium automation, accessibility scans, and BrowserStack AI agents. Tools that need local files or process state are marked unavailable on hosted MCP, including createTestCasesFromFile, Percy file helpers, and PRD upload. Hosted MCP cannot do Local Testing of localhost or VPN apps; that needs the local server. The vendor README states the server is under active development and supports a subset of the MCP spec, and that tool invocations can be non-deterministic. Official BrowserStackLocal is a tunnel binary (BrowserStackLocal --key), not a general test-management CLI. No official language management SDK was established from the opened API pages. iPaaS connectors were not opened. Computer-use viability is unassessed. API and MCP both score 7. API is the best path because it does not require an MCP client or LLM tool-calling and covers Automate build and session lifecycle over documented REST.

Best path today: api · Overall automatability: 7/10