Future Doors

Your camera roll is full of possible futures. Future Doors turns a saved opportunity into a source-checked career path that a person and an agent can build together.

The problem

People who are exploring more than one career direction constantly save internships, fellowships, challenges, research calls, and communities from Instagram, LinkedIn, and group chats. Those posts disappear into a camera roll. Even when someone returns to one, they still have to find the official page, interpret eligibility and dates, check whether it fits their schedule, and understand what the experience would actually add to their longer-term direction.

Most job products optimize for one role and one application. Future Doors is designed for people whose experience does not fit one label: a builder may also be a technical storyteller, researcher, or future education founder.

What it does

Future Doors turns scattered opportunity clues into a small, explainable plan:

A person shares a saved screenshot or link.
The agent finds the original official HTTPS source and stages the deadline, relevant rules, schedule, format, concrete outputs, and one unresolved fact for review.
The same profile can hold several possible futures. AI product, hardware storytelling, and learning, for example, produce genuinely different recommended work.
Source-checked cards are grouped as activities, projects, research, or credentials. Fit hearts represent verifiable alignment with the person's facts and preferences—not acceptance or hiring odds.
The person chooses up to three near-term moves. The plan shows timing, prerequisite relationships, schedule conflicts, and which goal-specific evidence each move can create.
Ineligible or unresolved opportunities remain visibly blocked. The agent can stage a separate source-backed path, but it cannot pretend that the original eligibility was restored.

The result is not another ranked list. It is a visual evidence deck: what to do, when it fits, what it can produce, and what is still missing.

Why WebMCP is essential

Future Doors is a shared stateful surface, not a chatbot wrapper. Through WebMCP, an agent can read the same profile, futures, saved opportunities, constraints, timing, and proof gaps that the person sees. It can then focus a route, compare constrained paths, check a saved post, surface one missing fact, or stage a source-backed change directly on the page instead of guessing through visual controls.

The live app registers 15 structured tools with document.modelContext.registerTool. Tool inputs are schema-validated, read-only operations are identified, externally sourced text is treated as untrusted, outputs are bounded, and invalid dates or IDs return explicit errors.

Most importantly, there is no WebMCP approval tool. An agent may prepare profile facts, an opportunity, a priority plan, an alternative route, or a work link, but the person must approve every consequential change in the interface. Source facts, agent inference, planned work, saved links, and verified proof are never presented as the same state.

That boundary is the product: the agent prepares; the person decides.

What is new

Future Doors combines three ideas that are normally separated:

a scrapbook for opportunities discovered outside job platforms;
official-source and eligibility checking by an agent;
a multi-future portfolio planner that shows how selected work can create reusable evidence.

It does not force a premature career choice and does not manufacture a success score. It helps someone make a bounded decision across several plausible futures while keeping every important claim inspectable.

How we built it

Next.js 16, React 19, and TypeScript
Browser-native WebMCP via document.modelContext.registerTool
15 scoped tools for reading, staging, comparing, simulating, and explaining paths
Vercel deployment with no account required for the judge demo
Vitest regression tests covering tool registration, route-aware recommendations, invalid inputs, source requirements, bounded outputs, rule mismatches, and human-only approval boundaries
Motion-based transitions that explain source checking, future-specific recommendations, card selection, and proof coverage

The current automated suite contains 38 passing tests. The production site has also been verified to expose all 15 tools and return a successful live get_path_snapshot invocation.

Try it

Open the live app and choose TRY THE EXAMPLE. Switch between Maya's AI, Hardware, and Learning futures to see different work. Add cards to the plan and watch the evidence coverage change. Open HOW IT WORKS for two judge-ready WebMCP prompts and the visible human-approval boundary.

No login or credentials are required.

Live app: https://future-doors-webmcp.vercel.app/
Public repository: https://github.com/aprilhan723/future-doors-webmcp