How it works.
MCP connects an AI assistant to tools and services. WebMCP gives websites a structured way to expose actions to an agent through the browser.
THE HUMAN OUTCOME
A task should survive the hand-off.
The assistant discovers available actions, checks what is authorised, completes permitted steps and reports the outcome.
Three parts. One completed task.
1. Describe the outcome
Tell the assistant what needs to happen, not which buttons to press.
2. Discover available actions
The website exposes defined tools and input requirements. Connected MCP services provide other permitted capabilities.
3. Prepare, approve and verify
The agent prepares the steps, requests approval for consequential actions, executes authorised work and reports what succeeded.
What is possible depends on browser support, the website’s exposed tools, connected services and permissions. We distinguish demonstrated capability from concept and roadmap.
Learn the terms