
Deepnatal
io.github.breezefeng由 GitHub 帳號 breezefeng 發布v0.1.2更新於 Oct 24, 2026
Natal charts cross-verified three ways. Refuses to answer rather than guess.
概覽
本機本命盤計算引擎,以三條獨立路徑交叉驗算出生星盤,路徑不一致時拒絕回傳結果。
- 可以做什麼
- 計算完整本命盤:十顆行星的星座、度數、宮位與逆行狀態,以及上升、天頂和十二宮頭;兩套引擎不一致時報錯而不回傳星盤。計算沒有出生時間時可知的部分,列出缺少項目與原因,並標示月亮星座不明確的情況。以純幾何路徑重算並裁定其他工具提供的上升星座數值。查詢某日期歷史時區資料的實際紀錄,包括當時是否實施日光節約時間。
- 適用情境
- 適合需要實際計算出的占星位置而非模型猜測的情況,尤其是出生地或年份涉及日光節約時間變動時,或需要判斷其他工具的上升星座是否因時區處理而出錯。
- 執行需求
- 以本機 Node.js 程序執行,透過 npx 從 deepnatal-mcp 這個 npm 套件啟動;未宣告帳號、API 金鑰或環境變數。全部在代理程式自身的程序中執行,不發出網路請求。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Deepnatal,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
deepnatal
A natal chart engine that cross-verifies every result three independent ways — and refuses to answer when it can't.
Ships as both a TypeScript library and an MCP server for Claude, Cursor, Codex, Cline and other agents.
Why this exists
Ask any astrology site for a birth chart and you get an answer. You almost never get a way to check it.
That matters more than it sounds, because the ascendant — the rising sign — moves about one degree every four minutes. Get the UTC instant wrong by an hour and it shifts by roughly half a sign. The result still looks completely plausible. Nothing about it announces that it's wrong.
And an hour is easy to lose. Taiwan observed daylight saving in 1945–1961 and again in 1974–1979. Mainland China did from 1986 to 1991. Japan did from 1948 to 1951. Plenty of tools quietly apply today's UTC offset to a 1988 birth date and hand back a chart that's off by exactly one hour.
This package is built around the opposite assumption: a result you can't check is not a result.
How it verifies
Three independent paths compute every chart. Disagreement is an error, not something to average away.
Typical agreement in practice is under one arcminute. Past tolerance, the call throws rather than returning a chart.
It also degrades honestly where the math genuinely breaks down:
- Above 66° latitude, Placidus houses are undefined. The engine falls back to whole-sign and tells you it did, instead of emitting silent nonsense.
- Without a birth time, the ascendant, midheaven and houses are simply omitted with reasons. No substituting noon. If the Moon changed sign that day — which happens on nearly half of all dates — that's reported too.
- Inside a DST transition gap, where a local time either doesn't exist or happens twice, it throws and asks you to disambiguate.
Verify it yourself
A README asserting "cross-verified" is worth nothing on its own. The suite ships inside the published package, so you can run it against the exact bytes you installed:
The 21 boundary cases are chosen to be the ones that break things, not the ones that are easy: China's 1986–1991 DST window and a same-year winter control, Taiwan 1979, Japan's 1948–1951 GHQ period, a time that never existed because a DST jump skipped it, and latitudes inside the Arctic Circle where Placidus is undefined.
The 11 external checks compare against values captured from astro-seek. Those exist because the three internal paths share assumptions — tropical zodiac, apparent geocentric positions, true ecliptic of date — and if an assumption itself were wrong, all three would agree on the same wrong answer. Only an outside source catches that.
CI runs the suite on Node 20/22/24 under a deliberately non-UTC system clock, so any accidental dependence on the host timezone fails loudly.
Quick start
Checking somebody else's answer
The reason this package exists. Pass a result from any other tool as claim:
Note degreesPerHour: 12.66. The usual rule of thumb says an hour is 15° — that's the Earth's rotation rate, not the ascendant's rate of travel along the ecliptic, and the two only agree near the equator. This library measures it per chart instead of assuming it.
MCP server
Claude Desktop — claude_desktop_config.json
Cursor — .cursor/mcp.json
Claude Code
Tools
Everything runs locally in the agent's own process. Birth data is about as sensitive as personal data gets, and none of it leaves the machine — there is no network call anywhere in this package.
Why let an agent call this instead of reasoning it out? Because a language model asked to work out a rising sign will produce a confident, plausible, wrong answer, and nothing downstream can detect that. This server returns computed facts and its own uncertainty. It never interprets.
中文說明
deepnatal 是一套會自我驗算的本命盤計算引擎——算不出來時它會拒絕回答,而不是給你一個看起來合理的答案。
上升星座每四分鐘移動約一度。時區算錯一小時,上升就偏掉大半個星座,而結果看起來完全正常,不會有任何地方提示你它是錯的。
而一小時很容易丟。台灣在 1945–1961、1974–1979 實施過夏令時間,中國大陸是 1986–1991,日本是 1948–1951。不少工具會把今天的時區偏移直接套到 1988 年的出生日期上。
三條獨立路徑同時計算,彼此不一致就報錯:
- 時區層 明確 IANA 時區 + Node ICU 歷史 tzdata,與座標反查時區交叉核對
- 行星層
circular-natal-horoscope-js×astronomy-engine兩套獨立演算法 - 上升層 純球面幾何第三條路(黃道與東方地平線交點),完全不依賴任何占星函式庫
實測吻合度在 1 角分以內。超過容許值就丟出錯誤,不回傳星盤。
verifyAscendant() 可以用來檢查任何其他工具算出的上升是否正確。它不猜原因——而是實際重算前後各一小時的上升點,如果對方的答案正好落在那裡,就能證明差異來自時區處理而非天文計算。
完整工具與說明見 deepnatal.com。
What this does not prove
Being straight about the boundaries is part of the point.
- Cross-verification catches implementation bugs and drift. It cannot prove the shared conventions are the right ones — all three paths assume the tropical zodiac, apparent geocentric positions and the true ecliptic of date.
- A reproducible chart says nothing about whether astrology predicts anything. This package computes positions. It makes no claim about meaning.
- Two tools can legitimately differ over house system, zodiac (tropical vs sidereal), ephemeris, node convention or rounding.
verify_ascendantdistinguishes a one-hour offset from these, but it doesn't declare a winner in matters of convention. - Wrong birth data still produces a technically correct chart of the wrong moment. Nothing downstream can rescue that.
Licensing
MIT. The dependency chain is deliberately kept clear for redistribution:
Notably not Swiss Ephemeris (sweph / swisseph). Its dual AGPL-3.0-or-later OR LGPL-3.0-or-later listing is misleading: the LGPL option requires purchasing an Astrodienst commercial license first. Without one you're on AGPL, which reaches network services and would require open-sourcing everything that touches it. That constraint is why most hosted astrology products can't ship their calculation layer as an open package — this one can.
Credits
This is the calculation layer of deepnatal.com, extracted verbatim — the site runs the same code that ships here, so any chart it produces can be reproduced locally against this package.
Issues and PRs: github.com/breezefeng/deepnatal
來源:README.md,提交 f9344a0
工具
0版本歷史
1- v0.1.2最新Sep 16, 2026
評分
已滿足適用檢查項的 50%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分