I vibe coded with GLM 5.2 an MCP server for Picat. This is a mean for LLMs to execute Picat code, which is very convenient to have in chat interfaces such as llama-ui (of llama.cpp), as it gives them the possibility to test the code they write, observe the syntax and logic errors, and iterate until the code is good.
The MCP server is available at
https://github.com/picat-lang/lib_ext/tree/main/use_Picat_with_LLMs . In the same directory, you will find another file that can help a lot the LLMs write correct Picat code, the Picat user guide in a very much shortened edition, in machine readable form (plain text). Even when you don't use the MCP server for Picat, you might want to pass this compact Picat reference to the LLM (regardless if it's chatgpt, one of the Antropic models or a local LLM), such that it is totally certain on the syntax, semantic and list of packages, predicates, functions and operators of Picat.
best regards,
CG