If you use AI agents - you still need to be able to take over manually. WebMCP is a standard proposal for a new in-browser AI control and it still requires accessibility, it can make us more efficient, but all users need to have the possibility to take control at any time.

Bogdan on A11y

WebMCP is a proposed browser-native standard that lets AI agents interact directly with web pages via predefined tools, removing the guesswork current AI automation relies on. While it promises efficiency gains — like context-aware filtering on e-commerce sites — it must not become a substitute for accessible web design. All parts of the loop (the webpage, the AI agent interface, and the WebMCP layer) must be accessible so users can take manual control at any time. The post warns against treating WebMCP as a 'conforming alternate version' that lets underlying pages remain inaccessible, and calls for disability advocates and accessibility specialists to be included in the standardization process. Practical barriers to testing the proposal — inaccessible demos and costly AI API tiers — are also flagged as issues that could exclude the very users whose feedback matters most.

WebMCP and the future of the agentic web: do not leave accessibility behind – Bogdan on Digital Accessibility (A11y)

Human in the loop needs accessibility of the whole loop

Standards in making need co-design or they may exclude

To conclude – WebMCP is not a replacement for accessible webpages