The three layers do not play the same role.
llms.txt is a context surface. It helps LLMs and agents find important pages, files, policies and resources on a site. It orients reading.
WebMCP is closer to structured action exposure. When relevant, it can help agents identify functions, capabilities or interactions more explicitly than a classic interface.
An agentic audit is a reliability verification. It checks whether the real site can be understood, navigated and operated correctly: HTML, DOM, visual rendering, accessibility tree, forms, CTAs, proof, internal links, governance and journeys.
The right hierarchy is therefore: llms.txt helps decide what to read, WebMCP can help decide what to execute, and the agentic audit verifies whether the whole system holds up in a real context.