LLMs are more efficient if they can henerate their insights by executing code (it is cheaper to exdcute determbistic code). As the size of context/attention is constraints and that we pay the generation per token (roughly words), a language that is succinct and dynamic is more approriate than more verbose alternative.

Clojure is a great tool for this. Its syntax is extremely simple and succinct.

This repos only focus on custom commands.

• Start a nrepl server at start
• Allow both cold execution of code and local server evaluation
• Adds many defaults libraries for performing typcial data transformation task:
  • Portal
  • malli
  • datascript
  • honeysql
  • data converter between json, yaml and edn
  • html generatiok with hiccup
  • http server and client
  • vega-lite support

We rely on the library to provide us the nrepl client and the necessary edting commands.

https://github.com/bhauman/clojure-mcp-light/blob/main/src/clojure_mcp_light/nrepl_client.clj