I built an MCP server that compresses and serves MySQL schema information.

When you ask an AI to write MySQL queries, you must give it the project’s schema. In very large projects a single database can have hundreds of tables. If you feed the whole schema dump to the AI, the context becomes too long and the request often fails.

To fix this, I thought an MCP server could return a neatly compressed schema so SQL generation works better. While learning MCP, I built a server that first fetches a table list and then fetches detailed information only for the tables you need.

More details in the blog post: https://medium.com/@shibayu36/i-built-an-mcp-server-that-com...