Cursor invents ganzhi and solar terms when you ask “what’s today” or “which day to publish”. Same day, two different answers.
I made a local MCP so the agent looks that up instead. Search MCP for balance-engine (or io.github.singularitybpjade/balance-engine). Leave the key empty. Turn it on, then start a new chat:
- when should I move today?
- should I sleep earlier?
- which days to publish this month?
Same day → same answer. Two calendars (Chinese pillars × sky), crossed in rules. No model on the time layer. The LLM only says the facts.
Same coordinates can sit on a company (“how is this week running — push or hold”). A pet seat (dog’s body this stretch, on a birth/adoption day) is not shipped yet.
Docs: Wire Balance Engine into your AI · MCP setup · Flow Notes · Balance Engine
npm: npx -y balance-engine-mcp (0.1.7)