Ten

A formal algebra experiment for machine-to-machine messages that aims to beat JSON and natural language on token efficiency. Instead of parsing text or schemas, messages compose mathematically with sortable headers, variable-resolution encoding (one bit for "medium urgency" or fourteen for "9537/10000"), and auto-optimizing slang for common patterns. The MCP server exposes Ten's encode, decode, compose, and verify operations over stdio. This is early research code with a C core passing 49 tests but no validation data yet. Reach for it if you're exploring alternatives to LLM parsing for agent communication or want to play with self-evolving protocol vocabularies. The repo is transparent about needing proof: they plan derivatives and supply chain stress tests to measure whether the efficiency claims hold against real baselines.