"What Happens Under the Hood" Section Clarifications - Can Developers Influence / Fine Tune the Tool Selection Process?

[princefishthrower]

Hi all, long time fan of anthropic and love how dev-friendly all your tools are. I'm thinking of taking a deep dive into MCP soon but I was wondering if I could get some clarifications here. In the "What Happens Under the Hood" documentation of the server-side quick start documentation, the following is listed:

1. The client sends your question to Claude
2. Claude analyzes the available tools and decides which one(s) to use
3. The client executes the chosen tool(s) through the MCP server
4. The results are sent back to Claude
5. Claude formulates a natural language response
6. The response is displayed to you!

I'm curious what effect we as MCP developers, if any, can have on this decision process, mainly steps 2. and 3. above. I can already imagine if I have two tools, one called addTwoDoubles and one called addTwoFloats, it could be tough for claude to determine which to use unless I explicitly say "treat them as floats" or "treat them as doubles". I guess what I am wondering if there would be a best practices way, to specify something like "when adding two decimals, always use the addTwoFloats tool unless the prompt explicitly specifies the two numbers as doubles", almost like a system prompt - or perhaps I could incorporate this into a MCP prompt itself? This is a fairly technical example of tools, but I could imagine the same issue comes up as soon as an MCP server has any considerable number of tools that are potentially overlapping in their use cases.

Is your feature request related to a problem? Please describe.
Not exactly a problem, but a clarification I would like before I start building.

Describe the solution you'd like
Additional notes in the documentation if it is indeed possible to influence steps 2 and 3, or if I'm totally breaking the whole point of MCP and that the model is supposed to make it's own decisions with MCP every time.

Describe alternatives you've considered
As I mentioned, ways of additional context or edge cases like "if this then", like a system prompt. Perhaps the server capabilities provide some sort of additional - by the way, the link to "Server Capabilities" in the README of the TypeScript SDK is broken

Additional context
I hope I made my point clear - if there already is a solution of what I am talking about with some examples, I'd be super happy!

Thanks!