Tweaksy

Inspiration

Accessibility settings are usually buried in menus and limited to a few predefined options. Yet every person’s needs are different—and those needs can change depending on the task, environment, or time of day.

We built Tweaksy to explore a more natural approach: instead of learning a website’s settings, people can simply describe what would help them. The website and AI then collaborate through WebMCP to adapt the experience safely.

What it does

Tweaksy lets people personalize a website using natural language.

A visitor can say:

“I’m color-blind and dyslexic. Please make this page easier for me to read.”

The AI uses tools exposed by the website to preview suitable changes, such as stronger contrast, color-independent cues, larger text, increased spacing, or a more readable typeface.

Visitors can also request a focus session that reduces visual noise, hides sponsored content, and deemphasizes distracting media.

Every change is applied directly at runtime—without rebuilding or redeploying the website—and can be previewed, approved, or discarded by the visitor.

How we built it

We created an interactive website that exposes ten accessibility and focus tools through WebMCP.

These tools allow an AI agent to:

• Inspect the page and its available capabilities
• Understand the visitor’s current personalization state
• Preview accessibility modes
• Adjust colors, typography, spacing, and content density
• Read page content aloud
• Start focus sessions
• Reduce distracting imagery

The website remains responsible for defining the safe actions that can be performed. The AI interprets the visitor’s natural-language request, selects the appropriate tools, and combines them into a personalized experience.

This creates a collaborative model: the visitor expresses the need, the AI translates it, and the website controls how the adaptation is applied.

Challenges we ran into

The biggest challenge was making the experience feel like a conversation rather than a collection of commands. Tool descriptions had to give the agent enough context to understand requests such as “help me focus” or “blue is difficult for me to see” without requiring the visitor to know technical terminology.

We also had to ensure that multiple adaptations worked together. For example, a dyslexia-friendly reading mode should remain active when a focus session hides advertisements and reduces imagery.

Accomplishments that we're proud of

We created a working, coherent experience in which an AI agent changes a live website through WebMCP—not through scripted chat responses or a simulated settings panel.

We are especially proud that:

• Visitors can describe their needs in ordinary language
• The agent can combine several tools in one interaction
• Adaptations happen immediately without rebuilding the page
• Changes are previewable and reversible
• Focus mode can hide real visual distractions
• The implementation demonstrates ten working WebMCP tools
• The same approach could help websites improve their WCAG and ADA readiness while serving a wider population with little work

Tweaksy turns accessibility from a fixed list of settings into an ongoing conversation between the person, the agent, and the website.

What we learned

WebMCP is especially valuable when a request is personal, contextual, and difficult to express through traditional controls.

A settings menu might offer “large text.” A person might instead say, “My eyes are tired, the text feels crowded, and the images are distracting.” An agent can understand the full intent and coordinate several website-defined capabilities to address it.

We also learned that WebMCP creates an important trust boundary. The agent does not need unrestricted control over the page. The website owner exposes specific, supported actions while the visitor remains in control of whether changes are kept.

Most importantly, we learned that AI does not have to replace visual interfaces. It can help each person reshape those interfaces around their own needs.

What's next for Tweaksy

Our next step is to develop Tweaksy’s WebMCP capabilities into a reusable accessibility protocol that website owners can adopt.

Instead of every site inventing different accessibility commands, a shared set of WebMCP tools could give agents a consistent way to request adaptations across the internet—while each website retains control over its design, implementation, and safety boundaries.

We also want to expand Tweaksy with persistent user preferences, additional accessibility profiles, improved screen-reader support, voice interaction, automated WCAG checks, and recommendations based on the content being viewed. You can find a Chrome Extention Github repo of Tweaksy as well that can work today on any website and persist the user prefernce but it would be less stable than each website exposing WebMCP

The long-term vision is simple: people should be able to bring their accessibility preferences with them everywhere on the internet.

Built With

• chatgpt-browser
• css
• html
• openai
• typescript
• vite
• vitest
• webmcp