MVP Scope

Walking skeletons, MoSCoW feature prioritization, the economic architecture, and what is explicitly deferred.

The MVP is a hypothesis in code form -- the version that collects maximum validated learning with minimum effort. It is not a "version 1.0."

Walking Skeletons

Two primary entry points, each representing a complete end-to-end flow through the system:

Skeleton A -- MCP Path (Day-to-Day Value)

User signs up on Trovella web
  --> connects MCP server to their AI app
  --> uses AI app as normal
  --> Trovella memory captures context
  --> next session in any AI app, memory is available
  --> user tries an MCP skill
  --> gets enhanced results

This skeleton validates the core hypothesis: cross-platform memory and curated skills make existing AI tools better without the user changing behavior. See Research & Intelligence for implementation.

Skeleton B -- Trovella App Path (Engagement + Fun)

User signs up
  --> onboarding captures basic preferences
  --> capability discovery shows what they can do
  --> user tries image generation
  --> creates a shareable image
  --> user explores RPG or preference recommendations
  --> returns to app

This skeleton validates the engagement hypothesis: guided experiences with immediate results create retention even before the MCP value becomes apparent.

Both skeletons share auth, user profile, preference engine, and memory as infrastructure.

Must Have (Launch)

Feature	Scope at Launch	Rationale
Cross-platform memory (MCP)	Hosted MCP server that stores and exposes memory across LLM platforms	Economic necessity -- heavy LLM work stays in subsidized apps. Also the strategic moat and data gravity engine.
MCP skills (5-10 curated)	Curated capabilities including early research tools, exposed through MCP	Immediate value in existing workflows without Trovella bearing token cost
Capability discovery system	Progressive revelation of what Trovella can do	This IS the product for users who don't know what to do with AI. Without it, features won't be found.
Image generation	Create pictures/paintings in creative scenarios using reference images. Google Gemini models.	Fastest "look what I just made" moment. High shareability. Proven demand. Primary hook.
Build-your-own adventure RPG	Single-player text-based RPG with AI storytelling and occasional generated images	Shares memory + image gen infra. Real motivated users (Kyle's daughters). Fun engagement hook.
Trovella chat (lightweight)	Minimal in-app LLM chat experience	Necessary at launch but not the front door. Fallback for users who want direct chat.
Preference engine (basic)	Tracks preferences, powers personalized recommendations	Makes the app feel like it knows you. Foundation for future network effects.
AI tutoring	Interactive lessons, practice exercises, skill progression	Directly addresses the core problem. Turns passive frustration into active learning.
Professional document generation	AI-assisted PowerPoint, Word, Excel from research results or prompts	High-value work output. Bridges "AI helped me think" to "AI helped me deliver."
Onboarding	Short preference-capture flow. LLM-powered for paid, non-LLM for free. Skippable.	First session must feel immediately different from opening a blank chat box.

Should Have (Shortly After Launch)

Feature	Rationale
Enhanced research tools	Starts as an MCP skill in the initial 5-10, then expands into a richer standalone experience. High value for financial analyst ICP but needs maturity first.

Could Have (Post-Launch, Pre-Growth)

Feature	Rationale
Social preferences	Link with friends/family for collective preference-based recommendations. Network effects need a network -- requires user base first.

Explicitly Deferred

These are not forgotten -- they have specific triggers for when they become relevant:

Item	Deferred Until	Details
Multi-tenant pricing tiers	Pre-beta	Free tier with no-LLM skills + paid tier targeting ~$8.99/month
Pricing model specifics	Pre-beta	See Go-to-Market for the framework
Beta launch plan	Phase 1	Paid closed beta with 20-30 users
REST public API	Post-traction	tRPC-OpenAPI bridge when external integrations are needed
Mobile app	Phase 2	React Native/Expo, shares tRPC API layer

The Economic Architecture

AI platforms subsidize consumer usage at roughly 12x ($200/month subscription yields ~$2,500/month API compute equivalent). Trovella cannot and should not compete with this subsidy. The feature scope reflects this constraint:

High-cost LLM work (deep research, multi-source synthesis, long conversations) stays in the user's AI platform via MCP
Low-cost targeted interactions (preference tracking, image generation prompts, structured data extraction) happen in Trovella where cost is controlled
Zero-cost enhancements (memory persistence, skill definitions, plan state tracking) are pure infrastructure with no per-request LLM cost

This is not a limitation -- it is the product strategy. See Problem & Product for the full architectural insight.

Development Approach

Walking skeleton first, then feature-complete vertical -- make one workflow genuinely good, not many workflows partially built
Budget allocation: 70% strategic features / 20% customer requests / 10% tech debt (pre-PMF)
Feature creep prevention: new ideas go to Phase 2 backlog; every addition requires removing something of equal effort
Solo founder timeline: 6-8 weeks to MVP with AI-assisted development (3-5x compression)

Problem & Product -- what each feature solves
Target Users -- why features are prioritized for financial analysts
Success Metrics -- how we measure whether MVP features deliver value
Go-to-Market -- beta and pricing strategy
Research & Intelligence -- technical implementation of the MCP path
Application -- technical implementation of the app path