RoomSplit

io.github.everyai-comv1.0.0更新於 Oct 4, 2026

Split rent and utilities by room, settle up and generate roommate agreements. Free.

已驗證Streamable HTTP可網頁執行Business & CommerceFinanceProductivity & Workflow

概覽

AI 產生的概覽

RoomSplit 是免費的遠端 MCP 伺服器，可依房間分攤房租與水電費、結清欠款並產生室友協議。

功能
RoomSplit 是託管的 MCP 端點，讓助理能依房間拆分家庭開銷、計算誰該付多少，並產生室友協議文件。登錄描述將其定位為免費的分攤房租與水電費、結清欠款的工具。清單中未列出工具，因此實際可呼叫的操作在此無法說明。
適用情境
適合合租家庭希望由助理協助依房間分攤房租與水電費、追蹤室友之間的帳目，或草擬共同居住協議的情境。它是託管服務，適合不想在本機執行程式的使用者。
執行需求
需要可連線至 streamable HTTP 端點的遠端 MCP 用戶端，網址為
安裝前請注意
清單宣告不需要身分驗證，因此任何能存取該端點的人都可能使用它；請避免傳送室友的個人、財務或身分資訊。隨本條目提供的 README 描述的是無關的占星專案，其中的安全與隱私說明不適用於 RoomSplit。產生的協議應視為草稿，而非法律意見。

安裝

在 SourceWeft 中

  1. 開啟 儀表板中的 RoomSplit，將其新增到工作區。
  2. 為需要使用其工具的對話啟用該服務。

Web executable，透過 Streamable HTTP。 遠端服務在工作區中設定後即可從網頁執行環境執行。

其他 MCP 客戶端

把它新增到你客戶端的 mcpServers 設定中。

{
  "mcpServers": {
    "room-split": {
      "type": "http",
      "url": "https://room-split.magicteams.ai/mcp"
    }
  }
}

README

Sahadeva

[CI]

A clean-room, Cloudflare-native research build for a transparent South Indian AI Jyotish workspace.

Run locally

bash
npm installnpm run dev

The Cloudflare Vite plugin serves the React application and Worker API together. POST /api/chart is deterministic. POST /api/interpret uses the configured Workers AI binding.

AI narration

Narration uses only Cloudflare's native AI binding and Cloudflare-hosted @cf/... models. The default is @cf/zai-org/glm-5.3-flash; AI_MODEL selects it in wrangler.jsonc, and non-@cf/ model names are rejected in favor of that safe default. No AI Gateway, third-party API, or provider token participates in this path. GET /api/ai/status exposes only non-secret readiness metadata, and successful narration identifies the Cloudflare model while keeping calculated evidence immutable.

The consultation surface adds deterministic topic judgments, house exploration, natal Panchanga, convention comparison, Prashna, timing fusion, rectification, and depth endpoints. Topic judgments share one evidence-ledger engine across web, MCP, reports, and the Expo mobile client. Opt-in snapshots store versioned ledgers and a hashed follow-up secret—not names, places, or raw birth input—and calibration remains descriptive rather than predictive.

Connect through MCP

After deployment, compatible clients can connect to:

text
https://your-worker.example/mcp

The Streamable HTTP endpoint exposes a high-signal catalog led by a comprehensive first-reading dossier. The returned profile reference lets later consultations answer focused questions from the verified natal context without repeating the dossier. It also supports chart reports, Prashna, timing, separate North Indian Ashtakoota and South Indian ten-Porutham compatibility, Panchanga, Muhurta and visual reports. Specialist tools remain directly callable and are documented through an MCP resource. Chart calculation does not require a language model. Public calls are rate-limited; optional scoped Bearer keys provide metering and vault integration. See MCP integration examples.

Clients can call assess_prediction_readiness before interpretation to distinguish implemented calculation from reviewed rules and calibrated prediction. Lal Kitab currently provides source-linked structural inspection. All of its material is retained, including sensitive topics and remedies; reviewed claims can later use graduated caution-led disclosure, while unreviewed personalized output remains withheld. See Prediction quality and MCP gap audit.

The source-only analyze_lal_kitab workflow now converts natal placements to Lal Kitab fixed houses and returns locators for all relevant planet-house sections without publishing their unreviewed claims. The complete integration inventory is in Lal Kitab integration map.

The web chart surfaces the same fixed-house map, source locators, retained-sensitive-material caution, and an independent calculation-quality audit. MCP clients can use the lal_kitab_consultation prompt or compose the equivalent tools from the published workflow resource.

For evidence-first use across AI clients, the MCP now publishes sahadeva://prediction-quality and the evidence_first_prediction prompt. Reviewed-rule search, rights-aware passage discovery, separate-tradition comparison, per-claim auditing, versioned outcome capture and conservative validation reporting are available as typed tools. These improve traceability and abstention; they do not guarantee prediction accuracy or convert unreviewed traditions into validated knowledge.

The primary consultation also supports a compact cross-tradition whole-person profile and consent-aware remedy selection. Its security contract is published at sahadeva://security; production profile references are keyed opaque identifiers, restricted knowledge remains server-side, and retrieved text is explicitly untrusted data. See MCP AI orchestration and security.

Workers AI has a limited daily free allocation. Sahadeva therefore describes hosted narration as optional and allowance-backed, not unlimited free inference.

Verify

bash
npm run checknpm testnpm run buildnpm run test:e2enpm run migrations:verifynpm run mobile:verify

Current status

The product interface and computation pipeline are functional. The Moon, Panchanga transitions, and Lahiri convention have high-precision reference fixtures; the application remains a research preview until independent Lagna, solar-event, complete-strength, and practitioner certification gates are satisfied. See RESEARCH_AND_ARCHITECTURE.md for the calculation boundary, topology, and validation gates.

Clean-room boundary

No third-party astrology engine is bundled. Repository research informed the capability map and test strategy only. Production algorithms must be derived from documented specifications and independently validated.

Continuous integration

Pushes and pull requests run the verify pipeline on Flare Actions (flare.yml): typecheck, unit tests, migration and secret checks, audit, build, bundle budget, Playwright e2e, and the Expo mobile verify. Run completions email all registered users.

License

MIT — see LICENSE.

來源：README.md，提交 684fa11

工具

0
工具後設資料尚未被收錄。

版本歷史

1
  1. v1.0.0最新Oct 4, 2026