Build a collaborative canvas where a human and an external AI agent share context through WebMCP. Human selections become temporary spoken keywords. The agent resolves those references, reads only the required content, communicates before acting, and leaves visible work on the canvas.
01
# Shared understanding The agent collaborates on the work, not on the webpage. ## Sequence 1. Resolve the human’s keyword. 2. Read the selected item details. 3. Communicate the intended action. 4. Make the change. 5. Report the result using the canonical keyword.
01| tool | purpose | status |
| canvas_list | Orient agent | Ready |
| canvas_resolve | Match speech to targets | Ready |
| canvas_read | Read content and geometry | Ready |
| canvas_capture | Inspect pixels | Ready |
| canvas_create | Create in a region | Ready |
| canvas_update | Edit content and comments | Ready |
Resolve first. Use the exact keyword. Announce before changing work. Keep the collaboration visible.