AgentPhone
Give an AI agent its own US or Canada number to call and text.
Run tests on real BrowserStack browsers and devices, debug failures and manage test cases in Claude.
Opens Agent Studio, where connecting is one click. The connector URL below works in any MCP client.
Paste it into any MCP client. Setup docs
The BrowserStack MCP server connects Claude to BrowserStack's testing platform. Anthropic's directory lists 37 tools for starting live sessions on real browsers and devices, setting up and running automated suites, debugging failures, scanning for accessibility issues and managing test cases and runs. It signs in with OAuth, and BrowserStack says data changes need your approval.
Verified connector
Listed by Anthropic as a partner connector in its Connectors Directory.
Connection checked by Agentman on .
Anthropic states this reflects the level of review a connector received, not a security audit.
Tool names from Anthropic's directory listing. This server requires sign-in, so we could not read tool descriptions or parameter schemas.
read, write and admin; BrowserStack does not document what each grants, or whether you can decline one on the consent screen.401, so schemas and annotations are not on this page.Yes. BrowserStack's MCP page states that an account is required, and its get-started guide notes a free trial is available. The remote server signs you in with OAuth, so you never paste a username or access key into Claude. Which tools return useful results depends on the BrowserStack products your account includes, such as Live, Automate, Test Management or Accessibility Testing.
No. BrowserStack's README lists no Local Testing support as a limitation of the remote server, so it cannot test apps behind a VPN, a firewall or on localhost. For those, BrowserStack points you to its local MCP server, which runs on your machine with Node.js 20.9 or later and authenticates with your BrowserStack username and access key rather than OAuth.
The MCP server brings BrowserStack's capabilities into your own AI assistant; the AI agents are built into BrowserStack's products. BrowserStack's FAQ describes its AI agents as handling specific tasks independently inside the product. Through MCP, Claude can still call some of them, such as low-code step generation with createLCASteps and self-healed selectors with fetchSelfHealedSelectors.
No. BrowserStack says that when it fetches a root cause analysis, the server returns the suggested fix as a proposal only and never applies code changes. Your assistant must show the suggestion and wait for your approval before editing files. The self-healing plan tool works the same way: BrowserStack describes prepareSelfHealingPlan as building an edit plan that does not modify files itself.
README.md and docs/TOOLS.md, retrieved 2026-11-18) · retrieved 2026-11-18POST https://mcp.browserstack.com/mcp returned 401 with a WWW-Authenticate challenge naming https://mcp.browserstack.com/.well-known/oauth-protected-resource, which lists https://mcp.browserstack.com as the authorization server and the scopes above; the path-append form returned 404 · retrieved 2026-11-18Connect BrowserStack once and your agents call these tools on their own: on a schedule, in a workflow, with nobody at the keyboard.