Saju MCP Server

Quick Verdict

Saju MCP Server is a community TypeScript server that exposes Korean Saju, or Four Pillars, calculations through MCP. Its tools cover a birth chart, interpretive fortune categories, two-person compatibility, solar/lunar calendar conversion, daily readings, ten-year cycles, period queries, and interpretation settings. The deterministic tables and rules can make it useful as a cultural-entertainment demo or an MCP implementation reference. They do not make fortune-telling scientific, and the output must not replace medical, legal, financial, relationship, or employment advice.

As checked on 2026-07-21, the repository manifest says MIT but the repository has no standalone LICENSE file. Its README advertises npm and one-line installation, yet there is no dependable npm package distribution to recommend. Install from reviewed source at a pinned commit instead. The code retains base chart inputs in memory for 60 minutes and major-cycle/yongsin data for 120 minutes, with hourly cleanup. That is still temporary retention and is not a user-facing deletion contract. Its HTTP example also needs authentication and restrictive CORS before network exposure.

Best For

• Developers studying MCP tool definitions, Zod validation, and a TypeScript server with stdio and HTTP entry points.
• Cultural or entertainment prototypes that use fictional or consented birth data and clearly label results as non-scientific.
• Technical users able to review source, dependencies, install scripts, cache behavior, and network configuration.
• Not suitable for professional advice, consequential decisions, or a public service that collects birth data without a privacy program.
• Not suitable for teams that cannot add authentication, narrow CORS, deletion controls, rate limits, and audit logging.

Key Features

• Integrated Saju analysis: Generates a Four Pillars chart and supports basic, fortune, yongsin, school-comparison, and methodology views.
• Compatibility: Compares two birth records and returns a traditional compatibility interpretation for entertainment use.
• Calendar conversion: Converts between supported solar and lunar dates with leap-month handling from local tables.
• Time-based readings: Provides daily, ten-year, yearly, monthly, hourly, and multi-year query modes.
• Deterministic local logic: Core chart and calendar operations use project rules and local data rather than a remote fortune API.
• Two transports: Can run as a local stdio child process or through the included HTTP server. The latter is an implementation option, not a secure hosted product.
• Interpretation settings: Lets a client inspect or change weighting presets, which may produce different narratives from the same birth input.

Use Cases

A low-risk use case is an isolated MCP lab: connect a test client, submit fictional records, inspect tool discovery and validation, and compare deterministic outputs. A cultural course can also use it to show how different traditional rule sets are encoded, provided every result is framed as entertainment or historical reference.

Public or multi-user use has a different risk profile. Birth date and time can contribute to identity and profiling. Although cache entries expire after 60 or 120 minutes and are cleaned hourly, TTL expiry is not user-controlled deletion. Do not duplicate those values into prompt history, analytics, ordinary logs, or an external model provider by default. Minimize fields, obtain consent, add explicit deletion, and keep the HTTP service behind an authenticated gateway.

Pricing

The source is available without a software fee; operators still pay for compute, maintenance, an MCP client, and any model service. A manifest value of MIT is not a substitute for the complete license text. Because the repository lacks a LICENSE file, commercial reuse, redistribution, and derivative deployment require additional confirmation from the maintainer.

Do not budget around an npm package merely because the README contains npm commands. This review treats source installation at a pinned commit as the available path. A third-party catalog, hosted template, or installer does not resolve package availability, software licensing, or security review.

Pros

• Consolidates chart, compatibility, calendar, and period queries into a small MCP tool set.
• Local rules and tables make core calculations reproducible and independent of a fortune-reading API.
• The TypeScript implementation is approachable for studying schemas, MCP dispatch, and transport choices.
• Includes both stdio and HTTP code, which is useful for learning why local and remote servers need different controls.
• Supports multiple interpretation modes without requiring the client to implement the underlying traditional rules.

Cons

• Fortune readings are not scientifically validated and can be overtrusted when presented in fluent language.
• README package claims do not match a dependable current npm distribution, so advertised quick-install flows may fail or drift.
• The manifest says MIT, but the missing repository LICENSE file leaves a licensing verification gap.
• Birth inputs remain in process memory for documented 60- or 120-minute TTLs, but there is no user-facing deletion contract.
• The HTTP path is unsafe for direct public exposure without authentication, restrictive CORS, TLS, limits, and audit controls.
• A remote shell installer can modify the machine and MCP client configuration; it should never be executed without review.

Alternatives

FAQ

Is Saju MCP Server a scientific prediction tool?

No. It encodes traditional Saju rules and produces interpretive text. Use it only for cultural reference, education, or entertainment, never as evidence for a consequential decision.

Can I install it directly from npm?

Do not rely on that path. The repository documentation contains npm commands, but a dependable current npm package could not be confirmed. Build reviewed source at a pinned commit and verify its lockfile and dependencies.

Is the project MIT licensed?

Its manifest declares MIT, but the repository does not contain a standalone license file with the full terms. Ask the maintainer for clarification before commercial use, modification, or redistribution and retain the answer as evidence.

Does it store birth data?

The implementation caches base chart data for 60 minutes and major-cycle/yongsin data for 120 minutes, with hourly cleanup. That is still retention even without a database. Minimize inputs, disable or clear the cache where needed, avoid ordinary logs, and define explicit deletion before accepting real data.

Can I expose its HTTP server directly to the internet?

No. Put it behind authentication and TLS, allow only specific origins and networks, add rate and size limits, redact logs, and audit calls. Keep it local or isolated until those controls exist.

Is the one-line installer safe?

Treat it as untrusted administrative code. Download and inspect it, verify the referenced commit and dependencies, understand configuration changes, and execute only the necessary commands manually rather than piping a moving remote branch into a shell.

Bottom Line

Saju MCP Server is most credible as a source-level MCP example and deterministic cultural-entertainment engine. It is not a scientific advisor. The unreliable npm distribution claim, missing full license file, 60/120-minute birth-data caches without user-controlled deletion, open HTTP/CORS posture, and remote installer risk rule out a blanket recommendation. If you experiment, pin and review the source, use fictional or minimal data, run over local stdio, and resolve licensing, authentication, deletion, and privacy controls first.