A divination engine that outputs prompts, not prophecies

What it does

Mingyu is an open-source toolkit covering a startling range of divination methods: Bazi, Ziwei Doushu, Western astrology, Liu Yao, Meihua, Qimen Dunjia, Da/Liu Ren, Tarot, Lenormand, temple fortune sticks, and date selection. You give it a birth time or a question; it does the chart calculation and returns structured data plus a ready-made prompt you can paste into any large language model for the reading itself.

The interesting bit

The architecture is the clever part: the hard, deterministic work — calendar conversions, true solar time, chart layouts — lives in a TypeScript core package (mingyu-core), while the fuzzy interpretive layer is delegated to whatever LLM you prefer. It’s a clean split between computation and interpretation, exposed via npm, a REST API, an MCP server, and an Agent Skill.

Key highlights

• mingyu-core on npm gives programmatic access: generate Bazi charts, cast Liu Yao hexagrams, draw Tarot spreads.
• Public REST API at aov.cc/api/v1 with OpenAPI spec; a single /prompt call returns both the computed chart and the interpretation task.
• MCP server runs locally via stdio (requests processed on your machine) or as a remote Streamable HTTP endpoint.
• Includes a 92-stick fortune system from the Sanshan Guowang temple in Jiexi, Guangdong, with original verses and stories preserved.
• An Android app can hand off generated charts to installed AI apps like ChatGPT or Claude, with API keys kept on-device.

Caveats

• The remote MCP endpoint has real limits: 512 KiB per request, no batching, and every request counts against Cloudflare quota — the docs explicitly warn against polling.
• AGPL-3.0-only licensing may matter if you’re building a product on top.
• Hobby project maintained by one person in their spare time; the “merit box” donations go to charity, which is charming but tells you about bus factor.

Verdict

Worth a look if you’re building anything AI-adjacent around Chinese metaphysics, or just want a well-structured prompt-engineering reference for domain-specific LLM tooling. If you were hoping for actual fortune-telling, the disclaimer is right there in the README.