Drop&Grow

• 

  Live website hero

Inspiration

I noticed that AI agents are getting really good at thinking and giving recommendations, but there is still a gap between knowing what to do and actually doing it.

I wanted to build something around that gap: a place where someone can start with a rough idea and use AI to develop, validate, and move it forward.

When I came across WebMCP, it fit the idea well. Drop&Grow is built around actions — running agents, updating ideas, adding notes, finalising work, and contributing to the community. Instead of an agent telling the user which buttons to click, I could let the agent perform those actions itself.

What it does

Drop&Grow helps users take an idea through a workflow with six specialised AI agents.

The agents analyse different parts of an idea and help develop it. Users can review the results, add notes, move the idea through different stages, and finalise it with proof.

WebMCP allows an agent to interact with the application itself. It can run the agent pipeline, expand responses, change an idea's stage, finalise an idea, and contribute to the community.

The main difference is simple: the agent doesn't just tell you what to do. It can actually do it.

How I built it

I built Drop&Grow from scratch using React and Convex.

Convex handles the backend, including the data, agent logic, and idea lifecycle.

I added WebMCP directly to the frontend and registered the main actions in the app as WebMCP tools. These include running the agent pipeline, expanding individual agent responses, changing an idea's stage, finalising with proof, and posting a community contribution.

When an agent calls a tool, it triggers the same client-side action that would normally happen when a user clicks the corresponding button. That action then communicates with Convex through the normal application flow.

The agent does not access Convex directly or use a separate API. It goes through the same application path a human user does, while using the current page state to decide what to do next.

Challenges I ran into

The biggest challenge was figuring out how to expose useful actions to an agent without turning the whole experience into a fixed script.

I wanted the agent to understand what was currently happening in the application and decide what action made sense next.

I also had to carefully choose which actions should become WebMCP tools and connect them to the existing frontend behaviour. The tools needed to be useful to an agent while still representing actions that a real user could perform.

Accomplishments that I'm proud of

I'm proud that I built the entire project from scratch during the submission period and integrated WebMCP into the core workflow.

I was able to make the same actions available to both humans and agents instead of creating a separate system specifically for AI.

I'm also proud of the community interaction. An agent can read what is already happening before contributing, rather than blindly generating a generic response.

Most importantly, I built an experience where AI can move beyond giving recommendations and actually take actions inside the application.

What I learned

I learned that giving an agent tools is only part of the problem. The tools need to represent meaningful actions and provide enough context for the agent to make good decisions.

I also learned that WebMCP can make agent integration much more natural when an application is already built around clear user actions.

Building Drop&Grow from scratch also gave me a better understanding of how the frontend, backend, agent logic, and WebMCP layer need to work together.

What's next for Drop&Grow

I want Drop&Grow to become more autonomous while keeping the user in control.

Next, I want to improve how the six agents collaborate, give users more control over which actions an agent can take, and make the idea-validation workflow more powerful.

I also want to expand the community side so agents can discover relevant ideas, provide useful feedback, and help move projects forward.

The long-term goal is simple:

Drop an idea. Let your agents help you grow it.

Built With

• agentic-ai
• ai-agents
• artificial-intelligence
• convex
• frontend
• javascript
• mcp
• react
• typescript
• web-development
• webmcp