Buddy — Your WebMCP Browser Agent

• 

  The Buddy website explaining how it work

• 

  Add to the cart after aproval

• 

  show the available options per the questions

• 

  Ask for approval before it proceed

• 

  Supported WebMCP from Shopify site explaining how you can do action and ask

Inspiration

I wanted to make WebMCP feel useful to everyday users, not just developers. The idea was to create a friendly browser companion that appears when a site supports WebMCP and lets people interact with it through simple chat or voice.

What it does

Buddy detects WebMCP-enabled websites, discovers the tools they expose, and lets users ask for actions in natural language. It can help search, navigate, and perform supported website actions while keeping higher-risk actions behind user approval.

How we built it

I built Buddy as a Chrome extension on top of WebMCP, using APIs like document modelContext.getTools() to discover site capabilities and executeTool() to run them. A lightweight agent service deployed on Railway uses the OpenAI API to interpret the user’s goal and select the next action from the tools exposed by the current website (which supports WebMCP). Buddy then executes that action through WebMCP in the browser while keeping permissions, risk checks, and user approvals under local control.

Challenges we ran into

The hardest part was handling different tool schemas and behaviors across websites while keeping execution safe and predictable. Voice interaction, browser permissions, argument validation, and action limits also required several rounds of debugging.

Accomplishments that we're proud of

I'm proud that Buddy became more than a technical demo. It provides a complete experience with WebMCP detection, natural-language interaction, voice support, approvals, activity feedback, and a friendly UI that works across supported websites.

What we learned

I learned how powerful WebMCP can be when websites expose structured capabilities instead of forcing agents to understand pages visually. I also learned that trust, clear feedback, and user control are just as important as the agent intelligence itself.

What's next for Buddy — Your WebMCP Browser Agent

Next, I want to expand Buddy and improve multi-step planning and voice interaction, and evolve it into a reusable browser-agent platform. Longer term, Buddy could become a SaaS product for individuals and teams, with persistent preferences, custom agent rules, shared workflows, analytics, and premium automation features across the WebMCP ecosystem.

Built With

• ai-agents
• browser-apis
• chrome
• github
• json-schema
• natural-language-processing
• railway
• rest-api
• speech-recognition
• speech-synthesis
• web-speech-api
• webmcp