Most discussions about AI in testing are stuck on chatbot demos: copying user stories into ChatGPT, asking for test cases, and pasting the code back into an IDE. That isn’t autonomous testing; it’s just interactive autocomplete.

True agentic testing requires giving AI models real "hands and eyes"—the ability to explore running applications, inspect logs, trigger test suites, and diagnose failures without a human babysitting every step. But building custom integrations to connect LLMs to browsers, devices, and test runners is fragile and hard to maintain.

That’s where the Model Context Protocol (MCP) changes the game. MCP provides an open standard for exposing tools, system state, and context to AI models.

In this session, I’ll share what I’ve learned building teswiz-mcp and integrating agentic workflows into test infrastructure. I’ll demonstrate how to expose test runners and inspection utilities as standardized MCP tools, show an agent diagnosing and triaging a failure across logs and UI states, and cover the practical guardrails you need to keep autonomous agents safe and reliable.