To-Do, powered by WebMCP

Every action below is a single handler. A person clicks it, or an AI agent calls it as a tool — same code, same result. The log shows who fired what.

WebMCP not detected

Running without WebMCP. The buttons and the “simulate agent” panel still work as a local demo. To activate the real browser API, open chrome://flags/#enable-webmcp-testing, set it to Enabled, and relaunch Chrome (needs Chrome 149+ for the origin trial, or the flag for local dev). Then test with the Model Context Tool Inspector extension.

Human view — click to act

Agent view — callable tools

Simulate an agent call

Human view — click to act

Agent view — callable tools

Activity log — both sides feed one stream