Build And Launch A Customer-Facing MCP
MCP turns the AI assistants your customers already use into a distribution surface for your product. Learn how to build one that survives contact with real users — from tool design to launch.
Led by Bharat Batra
What it's about
MCP turned every AI assistant your customers already use — ChatGPT, Claude, Copilot — into a distribution surface for your product. For teams that move early, it is a real growth and activation channel, not a checkbox integration.
But very few product teams can currently spec an MCP that can delight real users across key dimensions i.e reliability, tool naming, permission models and the failure states. It is a small, concrete skill with a short window of scarcity that AI forward PMs need to build.
This session is designed to help you learn exactly that skill: we cover whether you should build an MCP at all, the product design patterns that win, and the pitfalls that can kill most first attempts — grounded in what frontier tech companies ship everyday.
What you'll take away
- Where MCPs fit in the AI product stack — and whether you should build one at all.
- How to scope and name tools so an LLM can actually drive them and use your product as intended
- Shaping responses an assistant can act on — not just read back.
- Auth and permission flows that don't scare users off.
- The failure modes unique to MCPs: tool sprawl, permission mazes, and readability + how to design evals that catch them before they ship
Who should attend
AI PMs & Product Leaders
Speccing how your product should live inside AI assistants.
Founders
Weighing MCP vs CLI vs skills as a growth and distribution channel for your product.
Engineers & Tech Leads
About to build the MCP server your customers will actually touch.
Who'll be leading
Bharat Batra
Product Leader at Superhuman AI
Bharat Batra is a product leader at Superhuman AI. He has spent years building AI products like Superhuman Docs from zero to launch — learning firsthand what separates a product people rely on from one that quietly churns.
His conviction: with AI, the moat isn't features — it's craft. Getting code out is easy. Making the core loop work extremely well, and keeping it reliable in front of real users, is the grueling part most teams skip. He also teaches Luminary's AI PM cohort on shipping customer-facing agents.
Live on Zoom · 30 min
August 20 · 8:30 PM IST