tapfill

Inspiration

Forms are everywhere, but they were designed around clicks, typing, dropdowns, and rigid question flows.

We wondered: what if you could just talk to a form?

Instead of building another AI form product, we chose the form people already know and use: Google Forms. The goal was to make an existing web interface agent native without rebuilding it.

That became TapFill, a Chrome extension that adds a WebMCP powered voice interface to Google Forms, allowing an AI agent to understand, navigate, and fill forms through natural conversation.

What it does

TapFill turns an ordinary Google Form into a conversational interface.

Install the Chrome extension, open a Google Form, and activate TapFill. You can then simply talk to the form.

The agent can understand the form's structure, determine what information is needed, ask relevant questions, translate natural language into structured field values, fill the form, validate responses, and submit it.

The important part is that the form itself remains the interface. TapFill does not replace Google Forms with a new application. It adds an agent interface to the existing page.

A form built for humans can now be operated naturally by both humans and AI agents.

How we built it

TapFill is a Chrome extension that runs directly on the Google Forms page.

When a form is opened, the extension inspects its structure and creates a semantic representation of the available fields and actions. It then exposes those capabilities through WebMCP, allowing a compatible in browser agent to interact with the page through structured tools rather than trying to visually navigate or guess at DOM elements.

The extension provides tools for operations such as:

• reading the form schema and current state
• identifying unanswered fields
• setting field values
• validating responses
• submitting the form

On top of the WebMCP layer, we built a voice interface that handles the human side of the interaction. The agent turns conversational answers into structured tool calls while the underlying Google Form updates in real time.

The architecture is intentionally lightweight:

Chrome Extension → Google Forms → WebMCP Tools → AI Agent → Voice

This lets us demonstrate WebMCP on a widely used existing website rather than requiring a new application to be built from scratch.

Challenges we ran into

The biggest challenge was making an existing, third party web application agent friendly without rebuilding it.

Google Forms was never designed around an agent tool interface, so we had to bridge its existing UI and DOM structure with a clean, semantic WebMCP representation.

We also had to solve the harder problem behind "voice form filling": understanding conversational answers and reliably mapping them to the correct fields, while preserving the structured semantics and validation expected by the original form.

Finally, we had to make the experience feel like a conversation rather than speech to text controlling a webpage.

Accomplishments that we're proud of

We are proud that TapFill demonstrates a fundamentally different way of interacting with an existing website without replacing the website itself.

The most satisfying moment is watching a completely ordinary Google Form transform from a sequence of fields into a conversation.

Technically, we're particularly proud of using a Chrome extension to retrofit WebMCP capabilities onto an existing third party web application, showing how agent native interfaces can potentially be layered onto the web that already exists.

What we learned

We learned that the hard part of agentic interfaces is not simply getting an LLM to produce the right answer.

It is creating a reliable bridge between natural human intent and structured actions on the web.

We also learned that familiar interfaces are powerful. When the underlying Google Form stays visible and behaves exactly as expected, the leap from "AI demo" to "this could actually change how I use the web" happens almost instantly.

Most importantly, we learned that WebMCP can be interesting not only for websites built from scratch with agents in mind, but also as a mechanism for progressively making the existing web more agent accessible.

What's next for TapFill

Google Forms is just the beginning.

Our vision is to make existing web forms and workflows agent native, rather than forcing people to migrate to entirely new AI first applications.

TapFill could evolve from a voice layer for Google Forms into a broader compatibility layer for web applications, allowing agents to understand and operate interfaces that were never originally designed for them.

The web already has the interfaces. We want to give agents a way to use them.

Built With

• chrome
• codex
• crxjs
• gpt
• typescript
• vite
• webmcp