Great callout on the tool poisoning risk, Marcus. The OX Security findings are concerning because MCP is becoming the backbone for agent tool access across Claude Code, Cursor, and Gemini — and most developers are wiring in servers without auditing what tools they expose. The security model gets even more critical for commerce MCP servers where agents could handle real transactions and payment data. A compromised product catalog tool could inject malicious pricing or redirect purchases. The intersection of MCP security and ecommerce infrastructure is an important conversation that deserves more attention as agents start making real purchasing decisions.

Nice build walkthrough, Jangwook. The agent side of MCP is the harder half — most tutorials stop at the server, but discovering tools dynamically and composing them is where the real value is. Your TypeScript approach pairs well with what I have seen in production. When building an MCP server for ecommerce (covering search, filters, and comparison tools across a large product catalog), the agent-side tool discovery pattern you describe is exactly what makes it work: agents can query the catalog without knowing the structure ahead of time. For anyone building agentic shopping or product recommendation flows, the MCP pattern is a natural fit — the protocol handles the tool discovery that makes agentic commerce possible.

Excellent overview, Philippe. The key insight about standardizing context across LLMs is exactly why MCP adoption is accelerating — it solves the N*M integration problem that plagues every AI tool integration. One area where this matters in practice: commerce. Most ecommerce APIs return unstructured HTML or REST endpoints with bespoke auth. An MCP-native product catalog solves this by making millions of products available to any AI agent through a single protocol interface — search, filter, compare, all structured for machine consumption. I have been exploring this intersection and wrote about how MCP can serve as the missing infrastructure layer for AI agent shopping.