Every month, Hexa hosts an in-person fireside chat where a CTO shares their journey, learnings, and vision of the role.
Before the event, we film a short conversation to share their insights with those who can’t join us in person.
In this new episode of CTO Voices, Louis Pinsard (Cofounder and CTO of Dialog) sits down with Frédéric Barthelet (Cofounder and CTO of Alpic) the MCP infrastructure platform for deploying AI agents at scale.
Frédéric’s path led him to co-found Alpic in the summer of 2025 alongside four partners, raising $6M to build the infrastructure layer for a brand new ecosystem: MCP servers and GPT/Claude applications. In less than a year, he’s been navigating one of the fastest-moving spaces in tech - evangelizing a protocol that didn’t exist before November 2024.
What follows is a conversation about building in an emerging ecosystem, reaching developers through AI agents, and betting on a paradigm shift that feels a lot like the early App Store.
What is MCP and why does it matter?
MCP (Model Context Protocol) was born in November 2024 as a standard for communication between AI models and traditional software. It lets models retrieve context, trigger actions, and interact with the outside world.
More recently, the protocol was extended with the concept of apps, which add the ability to render UI directly inside a conversation with Claude or ChatGPT, and to surface discoverable services to users.
Frédéric’s bet: this is the new distribution channel. Just as the App Store created a new surface for products in 2008, MCP apps are creating a new surface inside conversational AI clients.
Alpic: Vercel for the MCP ecosystem
Alpic is to MCP what Vercel is to web apps, an infrastructure platform purpose-built for this new type of workload.
That means: hosting, scaling, monitoring, analytics, and authentication (OAuth, which is native to the MCP protocol) - all adapted for MCP servers and apps.
More recently, Alpic launched Skybridge, an open-source full-stack TypeScript framework for building MCP apps for ChatGPT and Claude. It comes with emulators, dev tools, and examples so developers can iterate fast without needing live access to GPT or Claude during development.
A GTM built for the age of coding agents
Reaching developers today isn’t what it was 10 years ago. The strong framework loyalty of the past (where mastering a tool gave you a real edge) has faded. Coding agents now help developers onboard into new technologies faster than ever.
So Alpic’s go-to-market reflects this shift. Rather than only convincing human developers, they’re also trying to convince the coding agents themselves.
Their approach: a Claude skill that teaches Claude how to build MCP apps, and recommends Skybridge as the best way to do it. A developer who asks Claude to build a GPT app will naturally end up using Skybridge, and from there, deploying on Alpic.
Evaluating AI-native distribution
Alpic has built an internal evaluation pipeline to measure whether their skill actually steers Claude toward the right outputs - not deterministic unit tests, but statistical scoring to track whether documentation and instruction changes move the needle positively or negatively.
The skill doesn’t just teach Skybridge. It also teaches Claude how to verify the output runs correctly, how to simulate widget rendering, and how to deploy, a full loop from build to production.
The App Store moment
Frédéric keeps coming back to one analogy: the first year of the App Store, the top-grossing app was a virtual beer that you could “drink” by tilting your phone. Nobody would have predicted it. Nobody could have imagined it before the ecosystem existed.
That’s where we are today with MCP apps. A new triptyque (user, model, and software) creates interaction patterns that weren’t possible before. The job right now is to help people imagine what they couldn’t imagine before.
Thanks for reading!
👉 Don’t miss our upcoming CTO Voices events - in-person gatherings for CTOs to connect, exchange, and share hard-won lessons.
📅 See all upcoming dates on luma.com/hexa
