An MCP server provides a list of tools, along with their parameters and descriptions, to an AI agent. For example it may have a tool like load_file
and a description saying how to use it. This description must be tokenized for the LLM to understand the tool.
It follows that using shorter descriptions is going to be faster—there is less text to tokenize and process for the LLM. For some words, like "the" and "a," tokens are required but the words add no extra meaning—so they can be removed.
We can write descriptions like "Load file from folder" instead of "Load the file from a folder." In removing these meaningless short words from my MCP tool descriptions, I reduced the token count from 2291 to 2125 tokens.
And given that local LLM inference is slower than most other processes on a computer, this probably was a worthwhile optimization. I noticed no quality degradation in the output; models are trained on text that has been written this way, so it might even work better.