NewsstandAI

The problem

AI agents are increasingly capable of researching and synthesizing information, but high-quality sources are often licensed, paywalled, or governed by access rules that a generic web search cannot represent.

NewsstandAI explores a simple question: what infrastructure would allow an agent to discover content across providers while respecting authentication, subscriptions, previews, and full-document entitlements?

Product architecture

The system separates four responsibilities:

• A provider interface normalizes search and document retrieval across source APIs.
• An Express backend owns authentication, users, subscriptions, and provider access.
• An MCP server exposes a small, stable tool surface to AI clients.
• A Next.js dashboard lets users understand providers, access, and account state.

OAuth 2.1 with PKCE authenticates remote MCP clients. Postgres and Drizzle persist users, clients, tokens, providers, and entitlements. Every document request resolves the user's access level before returning a preview or full content.

What I learned

The difficult part of an AI content product is rarely the chat surface. It is the system around the model: identity, provenance, rights, provider differences, failure modes, and a product model that gives each participant a reason to participate.

Team and execution

I architected and built the core product while leading a five-person team across delegated code contributions, product development, marketing, and publisher outreach. We used a spec → build → eval → ship workflow in Linear and maintained 156 automated tests across the monorepo.

The repository remains private while product, partner, and content-rights considerations are reviewed.