# Oraclebone [Health: Active]

**Category:** 🔮 Spirituality & Esoterica  
**Repository:** https://github.com/sapuyou45-bit/oraclebone  
**GitHub Stars:** 6  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/oraclebone

## Description
Audited tarot, I Ching, Xiao Liu Ren, Bazi (Four Pillars). Model never invents results.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `uvx` (confidence: high):

```json
"mcpServers": {
  "oraclebone": {
    "command": "uvx",
    "args": ["oraclebone"]
  }
}
```

## Documentation

## What the Oraclebone MCP server does

The Oraclebone MCP server provides a local MCP interface for four symbolic systems: tarot, I Ching, Xiao Liu Ren, and Bazi, also called Four Pillars. It is intended for reflection and symbolic reasoning rather than deterministic prediction. The server separates result generation from interpretation: audited local scripts perform the draw or cast, and the connected model works from that output.

Five MCP tools are included:

- `tarot_draw` draws cards using supported decks, spreads, and reversal settings.
- `iching_cast` produces a six-line hexagram, including primary and resulting hexagrams where applicable.
- `xiaoliuren_cast` casts from lunar-style numbers or a Gregorian-time fallback.
- `bazi_cast` creates a Bazi chart from a Gregorian birth datetime.
- `interpretation_template` returns a structured template for agent interpretation.

## How it works

The Oraclebone MCP server runs the underlying audited scripts locally and exposes their results to an MCP host. Normal operation uses system randomness. A seed can be supplied for tests and repeatable demonstrations, but seeded output is not presented as the default for real readings.

The scripts emit JSON containing enough information to inspect the method used. Traditional methods and their boundaries are documented by skill. Where an approximate method is available, the output includes a warning rather than representing it as traditional. This design prevents the model from silently inventing a card draw, hexagram, or other divination result.

The package also exposes a command-line interface named `ai-divination` and Python modules for direct use. The MCP layer is useful when an agent host should select and call the operation, while the CLI and Python API support scripts, tests, and workflows outside MCP.

## Setup and configuration

Install the package from PyPI with `pip install oraclebone`. The repository also supports installation from a checkout or editable installation during development. The MCP executable shipped by the package is `ai-divination-mcp`; configure that executable in an MCP-compatible host after installing the package.

The project documents use with Claude Desktop, Codex, Continue, and Cursor. It also provides agent-skill adapters for OpenAI/Codex, Claude Desktop or claude.ai project skills, Gemini CLI, and Cursor. These adapters route requests through the same `ai-divination` commands.

Bazi requires the optional `lunar-python` dependency. If only tarot, I Ching, or Xiao Liu Ren is needed, the other skills can be installed or copied independently. The shell installer targets `~/.claude/skills` by default; set `AI_SKILLS_DIR` to select another agent skill directory.

## Tools and capabilities

Tarot supports named decks and spreads, including major-arcana and three-card examples, with optional reversals. I Ching supports coin and yarrow methods. Xiao Liu Ren accepts number-based inputs and can use Gregorian time as a fallback. Each skill can be invoked through MCP, the CLI, or its underlying Python script, and the scripts return JSON.

The package can also install self-contained skill folders containing instructions, host metadata, scripts, and references. This is separate from using the MCP server, but it allows an agent to use the same audited operations through a skill workflow.

## Limitations and notes

Oraclebone does not establish scientific evidence for divination efficacy. Its value is procedural: it records how a symbolic result was generated and keeps interpretation distinct from generation. A seeded result is intended for reproducible demos and tests, not as a replacement for normal randomness.

Bazi is not available from the base installation unless the optional `lunar-python` extra is installed. The server is local and depends on the MCP host being able to launch the installed executable. The project does not claim that model interpretations are objective predictions; the model still supplies the interpretation after receiving the generated data.

_Full upstream README: https://allmcps.com/mcp/oraclebone/readme_