MCP in production: what nobody tells you before you start

MCP (Model Context Protocol) has been getting a lot of attention lately. And for good reason — it's a clean, open way to give AI models access to external tools and data. But if you're planning to put it in production, there are a few things worth knowing upfront. 1. Schema quality matters more than you think When an AI queries your database via MCP, it reads your schema. Vague column names, inconsistent conventions, undocumented relationships — the model will make wrong assumptions. Good schema design has always been important. With AI in the loop, it becomes critical. 2. Query volume will surprise you Once non-technical users can ask data questions freely, they will. A lot. We saw this building Conexor.io — teams that expected light usage ended up with 10x the query volume they planned for. Plan your connection pooling and rate limits accordingly. 3. Permissions need to be explicit Your MCP server should enforce read-only access by default. It sounds obvious, but it's easy to over-pr