Legend Saju

What Legend Saju MCP server does

Legend Saju MCP server connects MCP clients to an engine for source-traceable Eastern divination calculations. Its supported scope includes Saju and Myeongri, Zi Wei Dou Shu, Qimen Dunjia, Daliuren, Tieban, Korean naming analysis, dream interpretation, date selection, and question-based divination. The engine keeps different schools or calculation methods as separate results instead of collapsing them into one score.

The calculation path is deterministic and makes no model calls. Results can include selected methods, execution routes, claims, conflicts, timelines, evidence identifiers, and blocked systems. When a birth time is unknown, the engine preserves separate candidate charts, including alternative date-boundary handling, rather than silently substituting noon.

How it works

A host model first extracts usable details from the conversation and identifies an appropriate capability. The MCP tool then performs the calculation and returns structured data. The host model is responsible for turning that data into conversational explanations; the engine supplies the calculations, evidence, and boundaries.

Requests can control calculation depth and response format. Consumer output is intended for readable summaries, while evidence output exposes claims, calculation results, knowledge assets, source information, and related details. Debug output includes internal execution records. The TypeScript library also exposes a resolve entry point for applications that want to use the engine without an MCP client.

Setup and configuration

The hosted MCP endpoint is:

https://saju.tripsight.co.kr/mcp

The older Railway URL is documented as pointing to the same server. Claude Code can add the endpoint with its HTTP transport, and Codex can register it with its MCP URL command. Other HTTP-capable MCP clients can use a configuration containing the endpoint URL.

For local use, install Node.js 20 or newer, clone the repository, run npm ci, and build it with npm run build. The local server is ESM-based and runs through bin/legend-saju-mcp.js with STDIO. Applications can also install dependencies, build the project, and import dist/index.js directly.

Tools and capabilities

The Legend Saju MCP server provides purpose-specific tools including:

• legend_saju_read_fortune for general Saju and domain-specific fortune calculations.
• legend_saju_analyze_compatibility for comparing two people’s birth information.
• legend_saju_select_dates for explicitly requested candidate-date calculations.
• legend_saju_cast_divination for Qimen, Daliuren, I Ching, and other question-time or hexagram-based work.
• legend_saju_analyze_name for Korean name characters, naming evidence, and 81-number analysis.
• legend_saju_interpret_dream for cross-tradition dream motifs and conflicts.
• legend_saju_manifest and legend_saju_capabilities for inspecting available methods and engine coverage.
• legend_saju_run_methods for running selected schools or combined calculation plans.
• legend_saju_card_natal, legend_saju_card_fortune, legend_saju_card_compatibility, and legend_saju_card_timeline for structured card widgets.

The engine includes embedded knowledge and source data, so runtime operation does not require a remote database or hidden search service.

Limitations and notes

The hosted service applies limits on request size, requests per minute, and concurrent executions. The local and hosted versions process inputs without requiring OpenAI or Anthropic API keys, but the connected client still uses its own model session for conversation and explanation.

Card tools return calculated values from the server, while narrative text and scores are supplied by the host model and labeled accordingly. The project documents implementation boundaries for each tradition; the current registry should be checked with the manifest tool rather than relying on fixed capability counts. Results are calculation and evidence data, not necessarily finished fortune-telling prose.