Inspiration
Important decisions often get buried in long AI chats. Criteria drift, assumptions disappear, and it becomes hard to tell why a recommendation changed. We wanted a shared workspace where a person can see the decision model evolve instead of receiving an opaque answer.
What we built
Decision Workspace is a local-first decision canvas for humans and browser agents. It keeps options, weighted criteria, score-and-confidence assessments, evidence, uncertainties, constraints, and recommendations in one inspectable state.
The app exposes workflow-level WebMCP tools through navigator.modelContext. An agent can read the current workspace, ask one high-value interview question at a time, process an answer atomically, apply only grounded mappings, and adapt the next question to the new context. Free-text or uncertain answers remain explicitly unresolved instead of being forced into a score.
Users can explore temporary what-if scenarios without changing saved state. When the model is sufficiently grounded, the agent can create a readiness-checked Decision Brief with the leading option, drivers, remaining unknowns, and next verification action. The person remains in control of the final decision.
How we built it
We built the UI with Next.js, React, TypeScript, and Tailwind CSS. The canonical decision store is shared by the UI and the WebMCP adapter, so tool writes are immediately visible in the page and persisted locally on the device. Revision-aware writes and explicit question lifecycle states keep the interview consistent and auditable.
Challenges and what we learned
The hardest part was designing a small, coherent tool surface instead of exposing every UI control. We learned that an agent workflow needs explicit state reads, deterministic mapping policies, uncertainty preservation, non-destructive scenarios, and readiness gating. WebMCP works best when the browser page is the shared source of truth and the agent changes it through semantic workflow actions.
Built With
- local-first
- next.js
- react
- tailwind
- typescript
- vercel
- webmcp
Log in or sign up for Devpost to join the conversation.