WebMCP Side Panel - Keystroke Tool Selector

MEDIA PREVIEW

Overview & Result

Engineering leader Sarah Drasner built a prototype Chrome Side Panel extension utilizing WebMCP. By hooking into user input debounces, the extension queries Jev's typed choice API on each keystroke (under 40ms) to identify whether the user intends to search documentation, run DOM queries, inspect cookies, or call GitHub APIs, pre-rendering and arming the appropriate WebMCP tools before the user finishes typing.

How Jev fits in the loop

1. Monitors input state in Chrome Side Panel with a 150ms debounce window.
2. Passes active tab DOM context and current input string to Jev Choice API with a closed set of 12 registered WebMCP tools.
3. Jev returns the top-ranked tool and confidence score within 38ms without invoking full generative model overhead.
4. UI dynamically activates the tool chip, pre-fetching parameters so tool execution is instantaneous upon submission.

How to reproduce

1. Clone sdras/webmcp-panel and run npm install.
2. Build extension via npm run build and load unpacked directory into chrome://extensions.
3. Open any documentation page and toggle the Chrome Side Panel.
4. Type partial queries such as 'check storage' or 'find broken links' to observe sub-40ms tool activation badges.

Why this build matters

Proves that Jev's ultra-low latency enables UX paradigms that are impossible with standard generative models: ambient, real-time typing prediction and zero-latency tool routing directly in browser sidebars.

Reported performance

Cost: $0.0001 / query

Latency: 38ms / keystroke

Limitations

• Requires Chrome 114+ for Side Panel API support.
• Only predicts from pre-registered WebMCP tools within the extension sandbox.

Patterns