Servers Beyond 101 Good Practices, Design Choices and Their Consequences
summary
A concrete account of building MCP servers: an off‑the‑shelf Anthropics server sends 6–7 SQL queries per simple request, burns tokens, and once performed an unconfirmed ALTER TABLE that dropped data, prompting a rebuild; the replacement is domain‑specific, uses enums and explicit functions, is read‑only by default, sanitizes outputs to prevent prompt injection, and reduces calls and cost. Practical tooling notes: run a golden task suite across multiple model harnesses, use OASP guidance and NCPS scan, and require authentication and audited human approval for destructive actions.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:09Agentic AI rise and risks
- 4:40Simple MCP server limits
- 11:57Anthropic MCP server failure
- 17:25Real-world MCP server design
- 24:19MCP server security pitfalls
- 32:06Authentication and testing
- 35:39Scaling MCP servers
- 39:13MCP adoption maturity
- 39:57Q&A
related talks
25mYour Insecure MCP Server Won't Survive Production
02h 49mMCP Servers: Best Practices, Design Choices and Their Consequences
41mBending a Public MCP Server Without Breaking It
13mCloud MCP Servers Under Fire
17mBuilding Production-Ready MCP Servers: From AI Agents to Secure Enterprise APIs.
25mTools in your AI's Toolbox : An introduction to MCP Servers
28mSmarter MCP Servers: Using a Graph to Solve the Context Window Problem
01h 31mBuild AI-Powered Automation Tools: MCP Workshop for Developers & Engineers