Things that grid my gears: Folk creating whole new standards (TOON) vs using existing ones for the task at hand correctly (JSON). The example going virial on LinkedIn right now is not an apples to apples comparison. IF we treat JSON in the same manner then you can write the following VALID JSON that is just 26 tokens (less than TOONs 35 tokens): {"cols":["id","name","role"],"rows":[[1,"Alice","admin"],[2,"Bob","user"]]} Screenshot from toontools.vercel.app/playground below showing that TOON is 34.6% worse than just using JSON correctly for tabular data in an optimized way by thinking about the problem first. Alternatively if you want to maintain readability then its same as TOON (35 tokens each using this readable version - but if you are optimizing for tokens to save money then obviously you send the one liner above to your LLM. { "cols":["id","name","role"], "rows":[ [1,"Alice","admin"], [2,"Bob","user"] ] } #TOON #JSON #MCP #LLM #Tokens