Wenbu · 问卜

app.wenbuv1.3.1更新于 Oct 3, 2026

Free BaZi, I Ching, tarot, Zi Wei and bilingual learning tools with transparent methods. No API key.

已验证Streamable HTTP可网页运行Location & LifestyleKnowledge & Memory

概览

AI 生成的概览

问卜通过远程 MCP 端点提供免费的八字、易经、塔罗与紫微计算，以及可检索的资料库。

功能
该远程 MCP 服务器提供六个工具：calculate_bazi 按指定时区与换日规则排四柱，cast_iching 以三钱法或指定爻线起卦，draw_tarot 从 78 张牌中不放回抽取 1 张或 3 张，calculate_ziwei 依据已知当地出生时刻与传统性别参数排十二宫，search_library 与 read_library 用于检索和阅读问卜手册、符号条目与参考元数据。它只做计算与查阅，解读交由调用方的主机模型完成。locale 参数可切换中英文，部分传统名称仍保留中文。
适用场景
适合需要助手生成传统命盘或抽牌结果、或查阅相关术语与计算约定的场景；塔罗与易经不需要出生资料。更偏向学习与自我反思用途，而非预测；资料库工具可帮助解释与问卜公开约定保持一致。
运行要求
使用提供方的远程 Streamable HTTP 端点（/mcp 路径）；MCP 工具本身不需要 API 密钥、账号或本地运行时，解读由主机自带模型完成。需要能访问该服务的网络，并应显式设置 locale 参数。
安装前请注意
发送给工具的出生资料与问题会由提供方托管在 Cloudflare 的服务处理；README 说明 AI 路线会把消息与所选上下文转发给 DeepSeek，而 MCP 本身不调用该模型。资料库检索只返回外部链接与元数据，主机需自行打开页面后再引用内容。结果属于文化与反思性质，并非经科学验证的预测。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Wenbu · 问卜，将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Web executable，通过 Streamable HTTP。 远程服务在工作区中配置后即可从网页运行时运行。

其他 MCP 客户端

把它添加到你客户端的 mcpServers 配置中。

{
  "mcpServers": {
    "mcp": {
      "type": "http",
      "url": "https://wenbu.app/mcp"
    }
  }
}

README

Wenbu · 问卜

免费排盘、抽牌与查阅资料，把结果变成可以继续思考的问题。

Free charts, tarot and source-based learning for a question you want to explore.

中文网站 · English website · 知识手册 · Documentation

Open source / 开源接入

[Wenbu preview]

MCP quickstart · CLI · Agent Skill · Releases · Contribute

Official MCP Registry record · app.wenbu/mcp. Domain ownership is verified; this directory record does not imply endorsement or an installation. 官方目录登记已完成，可核对远程地址、版本与源码。

Connect remote MCP at https://wenbu.app/mcp, use the dependency-free CLI, or install the Skill from skills/wenbu/. Read the integration guide to choose a connection and run a reproducible example. The full application and integrations are MIT licensed. This is a public repository, not an anonymous-data or feedback export.

第一次用，从这里开始

你想做什么建议入口需要准备什么
还不知道怎么问Agent 对话选一个主题和目标，再修改生成的草稿；点击发送后才开始
围绕眼前一件事思考塔罗 / 易经一个具体问题，不需要出生资料
看出生资料对应的传统命盘八字 / 紫微公历生日与当地时间；八字还需时区，紫微必须知道时刻
弄懂术语与计算规则知识手册可以从入门内容读起，再核对结果里的计算约定

传统符号适合学习和自我反思。历法计算能否复核，与解释能否预测现实是两件事；问卜不把排盘结果当作人生结论。

有哪些功能

  • 八字：四柱、可见五行计数、未知时辰处理、午夜或子初换日选项，以及可选的近似真太阳时修正。五行数量不等于旺衰或喜用神。
  • 易经：三钱法起卦，保留自下而上的六爻、动爻、本卦与之卦。
  • 塔罗：完整 78 张牌，不放回抽取 1 张或 3 张，可选逆位；牌面为原创 AI 插画。
  • 紫微：十二宫与星曜，公开 iztro 的排盘约定；目前宫位和星曜名称保留中文。
  • Agent：使用 DeepSeek 调用实际工具、读取站内文章与精选外部资料、生成带来源的札记，并继续修改已有结果。研究范围限于资料目录。
  • 保存与接入：浏览器手记、会话导出、MCP、CLI、Skill 与 REST API。

免费与资料处理

计算、抽取、学习内容、手记和 MCP 无需注册或付费。单次 AI 解读每个网络每天最多 5 次，Agent 每天最多 12 回合，上海时间零点重置；仍受全站共享预算限制。AI 不可用时，独立工具仍可使用。详见免费额度。

会话与手记保存在当前浏览器，没有自动云同步。计算请求会发送至问卜的 Cloudflare 服务；使用 AI 时，消息和所选上下文会经该服务发送给 DeepSeek。导出文件可能包含私人资料，下载不等于自动分享，也不等于匿名化。详见隐私说明。

Start with a question

Use Wenbu Agent if you would like help framing a question. Choose a topic and an aim, edit the suggested draft, then send it. Choosing an option does not submit the draft for you.

For a question about the present, tarot and the I Ching need no birth details. BaZi and Zi Wei arrange traditional charts from birth information. The handbook explains how to begin and what the results mean.

The tools calculate or draw the symbols; AI can help discuss them. These are cultural and reflective practices, not scientifically established ways to predict a person's future. Visible element counts, for example, do not establish element strength or favorable elements.

Run locally / 本地运行

Requires Node.js 22.12 or later and npm. From this repository:

sh
npm cinpm run verifynpm run preview

Open localhost:8787. verify checks types, runs tests, builds the site and audits links and card assets. preview serves that build and the Worker API. Rebuild after changing static pages. npm run dev provides the faster Astro UI development server but does not supply Worker API routes.

AI is optional. To enable it locally, put server-only DEEPSEEK_API_KEY and QUOTA_SALT in an ignored .dev.vars file. No key belongs in browser code. See operations for setup, deployment and rollback.

Connect your own agent / 接入自己的 Agent

Use Streamable HTTP at https://wenbu.app/mcp. It exposes six tools:

ToolPurpose
calculate_baziFour Pillars with time-zone and day-boundary conventions
cast_ichingRandom three-coin cast or six supplied lines, bottom to top
draw_tarotOne or three cards from the 78-card deck, without replacement
calculate_ziweiTwelve palaces from a known local birth time and traditional sex parameter
search_librarySearch Wenbu guides, symbol notes and reference metadata
read_libraryRead a guide or symbol entry by its returned ID

MCP does not call DeepSeek or need your model key. Your host supplies its own interpretation. External reference search results are links and metadata; MCP has no read_reference tool, so the host must open those pages itself before citing their contents. Set locale explicitly to zh or en: MCP defaults to English, while REST and CLI calculations default to Chinese. Some traditional names remain in Chinese.

中文接入说明与可复制示例见 Agent 接入；英文说明见 Agent integrations。

Try the CLI

The checked-in CLI has no dependencies. This reproducible example uses no personal information and makes no model call:

sh
node public/wenbu.mjs --helpnode public/wenbu.mjs iching '{"lines":[7,7,7,7,7,7],"locale":"en"}'

It returns a kind: "iching" JSON result with the six supplied lines and no changing lines. For a new random draw:

sh
node public/wenbu.mjs tarot '{"count":3,"reversals":true,"locale":"en"}'

Keep personal inputs in a file or stdin rather than shell history: node public/wenbu.mjs bazi --file birth.json. The integration page supplies a complete synthetic file. WENBU_URL=http://127.0.0.1:8787 selects a local preview; WENBU_ANALYTICS=off opts out of coarse service analytics.

Read the Skill before adding it to your host. It does not grant access to other conversations or files. A chart's context export and an Agent API request have different schemas; do not pass an export directly to the agent command.

The built-in Agent

Open the workspace or use node public/wenbu.mjs agent --file request.json. This is a separate route that calls DeepSeek on Wenbu's account and requires the person's choice to share the supplied message and context. See the bilingual Agent protocol for a complete request and streaming examples.

The Agent can plan, calculate, read the curated library and available allowlisted pages, ask for missing details and write a report. It has no unrestricted web search or background execution. Reports and their citations are model-authored; a successful source read does not verify every interpretation. Follow-ups preserve the original cards or lines unless a new draw is requested.

Conversations and artifacts are saved in the browser. Requests send the selected context and a bounded recent history through Cloudflare to DeepSeek. Clearing browser storage removes local records; it does not delete provider records or reset quotas. The app does not promise end-to-end encryption or automatic cross-device sync.

Limits and architecture

AI allowanceCurrent configuration
Single reading5 attempts per network per Shanghai day
Agent conversation12 turns per network per Shanghai day
Work per Agent turnUp to 5 model calls, 12 tool executions and 120 seconds
Site-wide model budget1,000 attempts per day shared by both AI routes, with Agent use capped at 600

Shared networks can share an allowance. Failed or cancelled attempts may count. Calculators and MCP do not use the model budget, but request rate limits still apply. Visitors are not charged; the operator pays upstream and infrastructure costs.

Cloudflare Workers serves Astro static assets and API routes. A SQLite Durable Object manages AI quotas, D1 holds first-party product events and private feedback, private R2 preserves raw event archives, and rate-limit bindings protect the endpoints. The UI uses Astro 7 and React 19; calculations use pinned versions of lunar-typescript and iztro. Fonts are self-hosted. The optional model runs through DeepSeek's official API with requested model deepseek-v4-flash; responses record the model name reported by the provider.

Feedback is available from every public page and from reading/Agent results. Administrators can inspect linked usage history, triage feedback and download private NDJSON archives at /insights/. See storage, export and privacy details. Personal excerpts are shared only by an explicit choice in the feedback form.

Check and contribute

Public page changes are automatically sent to IndexNow after deployment, with a Cloudflare check every 15 minutes as a fallback. Submission receipts and retries are stored in D1; a receipt is not proof of indexing. See IndexNow operations.

See review evidence, calculation methodology and analytics definitions. Type checks, automated tests, browser checks, deployed smoke tests and actual search traffic are separate kinds of evidence. Software tests do not validate divinatory predictions.

Report a problem with reproducible steps, the selected convention and a synthetic example. Keep birth details, private questions and credentials out of public issues.

MIT license. Third-party packages retain their own licenses.

来源：README.md，提交 ce2fe06

工具

0
工具元数据尚未被收录。

版本历史

1
  1. v1.3.1最新Oct 3, 2026