Dodolink

Inspiration Well I was inspired by webmcp that list 10 people for the award that's wonderful and I believe I'd be one of the winner, that's why I joined.

What it does DodoLink Agent Shopping gives browser agents a structured way to help shoppers navigate a real Nigerian marketplace. Instead of scraping repetitive listing cards and guessing which actions are safe, an agent can call typed WebMCP tools to search products, discover local stores, compare prices, and prepare a purchase plan.

How we built it I built DodoLink Agent Shopping as a web-only WebMCP experience connected to DodoLink’s real public marketplace data.

The page exposes four structured, read-only tools:

search_products discover_stores compare_products prepare_purchase_plan Each tool is registered imperatively with document.modelContext.registerTool, using a JSON input schema and an asynchronous execute function. The tools return real product and store information, including prices, stock status, seller details, and review links.

I also added a manual fallback so the experience still works in a normal browser when WebMCP is unavailable. Both the manual interface and the WebMCP tools use the same marketplace search and purchase-planning logic.

The system intentionally does not expose payment, checkout, or seller-messaging tools. The agent can discover and organize options, but the human must review the products and decide whether to continue.

Challenges we ran into An agent could describe a product, but it could not reliably distinguish listing IDs, price limits, stock, seller context, and review links. The WebMCP tools make those capabilities explicit and return bounded public data. The plan tool then turns agent work into a visible human approval step.

Accomplishments that we're proud of We are proud that DodoLink Agent Shopping is connected to real marketplace data instead of using a mock catalog.

We built a complete experience where an agent can:

Understand a natural-language shopping request. Search real DodoLink products. Discover relevant stores. Compare product options. Prepare a shortlist and purchase plan. Hand the decision back to the human shopper. We are especially proud of the safety boundary. The agent can help with discovery and research, but it cannot spend money, check out, or send messages to sellers. This keeps the agent useful while ensuring the human remains in control of important decisions.

We also created a separate public and sanitized open-source repository so judges can inspect the WebMCP implementation without exposing DodoLink’s private production infrastructure.

What we learned We learned that WebMCP is most useful when an application exposes focused capabilities instead of allowing an agent to guess its way through a user interface.

For a marketplace, structured tools are much more reliable than page scraping. A tool can clearly define the product query, price limit, product IDs, and expected response format.

We also learned that WebMCP tools need strong safety boundaries. Marketplace listings and seller descriptions should be treated as untrusted content, while actions such as payment and messaging should remain under human control.

Another important lesson was that WebMCP is page-scoped and still experimental. A user must open the WebMCP-enabled page in a compatible browser before the agent can discover and use its tools. Because of this, we made the manual experience work as a fallback too.

What's next for Dodolink Next, we want to make the agent experience more useful for complex shopping decisions.

We plan to improve product filtering with more detailed category, location, availability, and seller-reputation criteria. We also want to make comparisons more informative by including delivery options, product attributes, review summaries, and clearer trade-offs between sellers.

We would like to improve the human handoff so shoppers can save a shortlist, share it with someone else, open a seller profile, and continue manually when they are ready.

As WebMCP develops, we also plan to test DodoLink with more browser agents and refine the tool schemas based on real usage. Our long-term goal is to make DodoLink a marketplace where agents are genuinely helpful while people remain informed and in control.

Built With

• api
• css
• javascript
• node.js
• rest