Writing the descriptions in an MCP server is like writing prompts for the tools. You have to explain what the tool does, when to use it, and what the parameters are and how to use them.
Essentially it is like coding in natural language, and it can easily lead to bloated prompts when running an AI agent. I wanted to add support for relative dates, like today and yesterday, so I ended up requiring about 4 new tool calls, each with detailed prompts.
Now my agent is more capable—it can resolve things like "today's post"—but it is also dealing with over a thousand tokens of prompts. It is a good thing I can get by with a smaller model (Gemma 4 E4B) as performance problems would become apparent otherwise.
It does seem like there is a tradeoff between agentic AI ability and prompt processing—the fewer instructions an agent has, and the fewer tools, the faster it is, and the less capable. In any case, there is likely a point where the MCP server has enough features to suffice, and adding more will just slow things down and make it more complicated and unreliable.