Branchiverse

Inspiration

Most AI decision tools end with an answer. Real futures are not answers: they are living systems of choices, assumptions, turning points, and consequences. We wanted a shared space where a person could bring their values and lived context while an agent contributed imagination, structure, and overlooked possibilities.

What it does

Branchiverse turns an open-ended scenario into an interactive branching universe. A person manipulates branches and events directly on the canvas while an external AI agent inspects the same state and uses structured WebMCP tools to add branches, edit events, explore what-ifs, duplicate paths, collapse branches, and compare outcomes. The demo follows a personal future: moving to Lisbon or staying in Montréal. After inspecting both paths, the agent discovers a third option—a seasonal, transatlantic life—and adds connected events that make the original decision reversible. The final artifact is not an AI answer. It is a future model that neither participant would have created alone.

How we built it

Branchiverse is a responsive React application built with Vinext, TypeScript, and CSS. Its shared universe is structured branch and event data. The page registers eleven tools through document.modelContext, allowing compatible external agents to act on that state while React immediately reflects changes on the visual canvas. The WebMCP tools are create_universe, add_branch, add_event, edit_event, remove_event, branch_from_event, explore_what_if, compare_branches, duplicate_branch, collapse_branch, and get_universe_state. WebMCP is a progressive enhancement: the canvas remains usable when no agent is connected.

Why WebMCP matters

Without WebMCP, the agent could only describe suggested changes in chat and ask the human to reproduce them. With WebMCP, the agent participates in the same creative surface. The canvas—not the chat transcript—becomes the shared artifact.

Challenges

The central design challenge was keeping collaboration legible. Agent actions needed to feel immediate without obscuring human authorship. We addressed this with provenance labels, live activity, distinct branch colors, tool-use summaries, and a focused action surface.

Accomplishments

• A complete eleven-tool WebMCP surface
• Real-time agent mutations reflected on the canvas
• A polished, responsive scenario that communicates the concept immediately
• Graceful behavior with and without WebMCP support
• A focused demo flow under three minutes

What we learned

The most compelling agent experiences are not necessarily autonomous. Branchiverse works because human intention and agent exploration remain visible at the same time. Structured browser tools let the agent become a collaborator inside an existing interface rather than replacing that interface with chat.

What's next

Future versions could add persistent universes, multiplayer sessions, uncertainty ranges, evidence attached to assumptions, replayable history, and templates for business futures, alternate history, fictional worlds, and civic planning.

Built With

• css
• lucide
• react
• typescript
• vinext
• vite
• webmcp