I wonder if you would need to expose each command as its own tool. Might be able to expose a single tool with a description and context on how to discover and call commands therein?
For how to work with command line tools built with System.CommandLine e.g. built-in commands, conventions for args and options?
Some fun building an MCP server to expose tools for querying any search system, judge candidate results and calculate relevancy metrics like NDCG and ERR:
www.youtube.com/watch?v=vLhp...
It inferred tools to call and how to build queries
#search #relevancy #evaluation #metrics #mcp
May want to take a look at Brave and Arc
Totally! The key is framing improvements in terms of business value- revenue impact, conversion rates, retention. "Better search relevance" isn’t compelling; "X% increase in purchases due to Y% improved relevance" is, with experimental results to back it. Make it about outcomes.
🚀 Introducing Search Pioneer Ranking CLI – An open source Learning to Rank (LTR) command line tool for training rankers with LightGBM and FastTree:
github.com/searchpionee...
✅ Train rankers with LightGBM & FastTree
✅ Calculate feature importance using PFI
✅ Split and Fold data sets for training
