PrizePilot WebMCP

A human-controlled WebMCP workspace that helps people and agents organize verified funding applications without losing oversight.

Comment

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for PrizePilot WebMCP

Inspiration

Funding and hackathon applications are repetitive, deadline-driven, and easy to lose track of. Existing trackers are passive: they store records, while browser agents often rely on brittle screen scraping. PrizePilot WebMCP explores a better contract—structured tools that let a person and an agent work from the same verified record without giving away control.

What it does

PrizePilot WebMCP turns opportunities into a prioritized, evidence-led queue. The live page exposes four WebMCP tools:

  • search_opportunities finds verified local records by project, organizer, tag, or status.
  • build_application_plan produces a five-step, evidence-led checklist without changing data.
  • stage_status_change proposes a local status update for human review.
  • record_verified_submission requires an HTTPS receipt before it can stage Submitted status.

Every tool call is visible in the activity log. Any state-changing request opens a clear confirmation card. Nothing is submitted to an external form automatically.

How we built it

The prototype is a responsive React and TypeScript application built with Next.js-compatible components and vinext, hosted on ChatGPT Sites. It registers tools through document.modelContext.registerTool with bounded JSON schemas, length limits, read-only annotations, and explicit human-confirmation semantics. Demo records are stored locally in the browser, so the experience remains useful without an account or external database.

A progressive fallback detects browsers without WebMCP and keeps the full dashboard usable. The same interface supports normal manual interaction, a safe built-in agent-plan demo, and real WebMCP calls from compatible agents.

Challenges we ran into

WebMCP is an emerging standard, so we designed for uncertain browser support and made every capability progressively enhanced. The hardest product decision was not how much the agent could do, but where it should stop. We treated submission claims and consequential status changes as evidence-backed proposals rather than silent automation.

Accomplishments that we're proud of

  • A working public prototype with four registered WebMCP tools.
  • A shared visual activity log for people and agents.
  • Receipt-gated submission records and visible confirmation before local mutations.
  • A complete no-login fallback that preserves the product experience.
  • Clear separation between organizing work and submitting external applications.

What we learned

Structured page tools are more reliable and understandable than screen scraping. More importantly, WebMCP can make agent behavior easier to inspect: the interface, schemas, annotations, and confirmation state become a shared contract between the site, the user, and the agent.

What's next for PrizePilot WebMCP

Next steps are official-source synchronization, reusable evidence libraries, team review, receipt exports, deadline alerts, and connectors that can prepare—but never silently submit—applications across funding platforms.

Built With

Share this project:

Updates