# asktian MCP server [![CI](https://github.com/douglasgan/asktian-mcp/actions/workflows/ci.yml/badge.svg)](https://github.com/douglasgan/asktian-mcp/actions/workflows/ci.yml) [![npm](https://img.shields.io/npm/v/@asktian/mcp-server)](https://www.npmjs.com/package/@asktian/mcp-server) [![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE) [![asktian MCP server](https://glama.ai/mcp/servers/douglasgan/asktian-mcp/badges/score.svg)](https://glama.ai/mcp/servers/douglasgan/asktian-mcp) > **Chinese metaphysics inside your AI assistant.** > 不知道就问天 · Ask tian when you don't know. Adds 4000 years of Chinese metaphysical traditions (bazi / 八字, qimen / 奇门, five elements, daily 干支 energy) as callable tools for any AI agent that speaks the **Model Context Protocol** — Claude Desktop, Claude Code, Cursor, Windsurf, Zed, and any MCP-compatible client. When you ask your AI assistant a timing, person, or decision question, it can quietly consult asktian before answering. **The decision layer for AI agents.** Agents already plug in search, memory, and payments — askTIAN is the layer that gives them a *stance*: a specific, falsifiable call instead of *"it depends."* → [Why it sells](docs/POSITIONING.md) · [2-min quickstart](docs/quickstart.md) > ### ✨ Real readings, **no key needed.** > > The tools call the live askTIAN backend and return **real** results — daily almanac, fate > compatibility, name analysis, day energy — straight out of the box. The read endpoints are > **public and rate-limited** (~300 requests / 15 min). An **optional** `ASKTIAN_API_KEY` > ([api.asktian.com](https://api.asktian.com)) raises the limit and unlocks the **premium > `tian.*` reading tier**. `market_read` is a deterministic local novelty signal; `best_time` > returns a guided preview (no backend endpoint yet). ``` You: "I have a hard conversation with my boss tomorrow at 3pm. Should I move it?" Claude (with asktian MCP): → calls asktian_best_time_for_action({ birthdate, action: "difficult_conversation" }) → "Friday 10am scores much higher. Tomorrow 3pm is your clash hour. Want me to draft a message to reschedule?" ``` --- ## Install ```bash npm install -g @asktian/mcp-server ``` Then add to your client's MCP config: ### Claude Desktop Edit `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%\Claude\claude_desktop_config.json` (Windows): ```json { "mcpServers": { "asktian": { "command": "asktian-mcp", "env": { "ASKTIAN_API_KEY": "at_live_xxxxxxxxxxxxxxxx" } } } } ``` Restart Claude Desktop. Look for the 🔌 icon — asktian should appear with 6 tools. ### Cursor Add to `~/.cursor/mcp.json`: ```json { "mcpServers": { "asktian": { "command": "asktian-mcp", "env": { "ASKTIAN_API_KEY": "at_live_xxxxxxxxxxxxxxxx" } } } } ``` ### Windsurf Add to `~/.codeium/windsurf/mcp_config.json` — same shape as above. ### Zed Add to your Zed settings under `assistant.mcp_servers` — same command + env. ### Any other MCP client Run `asktian-mcp` as a subprocess; it speaks JSON-RPC over stdio per the [MCP spec](https://spec.modelcontextprotocol.io). ### Remote / hosted (streamable-HTTP) For LangChain, LlamaIndex, Fetch.ai, or any agent that connects to a hosted MCP endpoint, run the server in HTTP mode instead of stdio: ```bash ASKTIAN_MCP_HTTP=1 PORT=8080 npx -y @asktian/mcp-server # → streamable-HTTP on :8080/mcp (health check: GET :8080/health) ``` Hosted at **`https://mcp.asktian.com/mcp`**. Pass your key as a Bearer header: `Authorization: Bearer YOUR_ASKTIAN_API_KEY`. ### LangChain & LlamaIndex Both have native MCP adapters — askTIAN works with **zero integration code**. Copy-paste snippets in **[docs/langchain-llamaindex.md](docs/langchain-llamaindex.md)**. --- ## API key **The key is optional — readings are real either way.** | | 🆓 **No key** (default) | 🔑 **With key** | |---|---|---| | Setup | nothing — works instantly | get a key at [api.asktian.com](https://api.asktian.com) | | Readings | **real** — live api.asktian.com backend (daily almanac, compatibility, name analysis, day energy) | same real readings | | Rate limit | ~**300 requests / 15 min** (per IP) | **higher limits** | | Premium `tian.*` tier | — | **unlocked** | | Cost | **free** | per the key's plan | **To use a key** (higher limits + premium tier), set `ASKTIAN_API_KEY` in your client config: ```jsonc { "mcpServers": { "asktian": { "command": "npx", "args": ["-y", "@asktian/mcp-server"], "env": { "ASKTIAN_API_KEY": "at_live_..." } // ← omit this line to stay in local mode } } } ``` Hosted (streamable-HTTP)? Pass it as a header instead: `Authorization: Bearer at_live_...`. --- ## The 6 tools ### 1. `asktian_daily_reading` Personalized daily energy reading for a person. ``` input: { birthdate: "1992-05-15", birth_hour?: "14:30", gender?: "male"|"female"|"any" } output: archetype (one of 8 trigrams), today's stem-branch energy, favorable colors / direction / hours, headline + body advice ``` **Use when:** user asks how today will be, what colors to wear, what direction to face their desk, daily guidance. ### 2. `asktian_compatibility` Fate compatibility between two people. Returns qualitative label first (e.g. *"互补型 Complementary"*), then numeric score (hidden if <60 per asktian design principles — low compat should never feel like rejection). ``` input: { person_a_birthdate, person_b_birthdate, dimension?: "love"|"career"|"friend"|"general" } output: qualitative label, category, element flow, today's advice, score (with should_show_score flag) ``` **Use when:** "will this person and I work", "compatibility check", "is this a good match". ### 3. `asktian_best_time_for_action` The killer tool. Find the most auspicious time windows for a specific action over the next N days. ``` input: { birthdate, action: "difficult_conversation"|"negotiation"|"launch"|... , range_days?: 7 } output: top 3 best windows (date + hour + score + reason), windows to avoid ``` **Use when:** "when should I do X", "should I move this meeting", "is tomorrow a good day to launch", "when should I have the hard talk". ### 4. `asktian_today_energy` General energy of the day — no birthdate needed. ``` input: { date?: "YYYY-MM-DD" } (defaults to today) output: 干支 stem+branch, dominant 5-element character, description ``` **Use when:** AI wants to add cosmic context to a generic suggestion without needing the user's birthdate. ### 5. `asktian_name_analysis` Quick energetic profile of a name (姓名学). ``` input: { name, language?: "en"|"zh"|"auto" } output: dominant element guess, tone, one-liner ``` **Use when:** discussing baby names, company names, or "what kind of person is X" when birthdate unknown. ### 6. `asktian_market_read` A Chinese-metaphysics signal on a binary prediction-market question (Polymarket/Kalshi style). ``` input: { question, resolve_date?: YYYY-MM-DD, subject_birthdate?: YYYY-MM-DD } output: { signal: { lean: yes|no|neutral, score 5–95, confidence }, reasoning, disclaimer } ``` **Use when:** a user or a trading agent wants an *uncorrelated, for-fun* read on a market — "will X happen by date Y". The value isn't prediction; it's a **deterministic signal that doesn't read the same news every LLM reads**, published so the calls can be scored over time. ⚠️ **Entertainment / falsifiable ritual — NOT financial advice.** Every response says so, and the tool is built to be presented as a novelty, never as a bet recommendation. --- ## Design principles the tools follow asktian is opinionated. The MCP tools surface these constraints to your AI agent through the `note_for_ai` field on each response. Notable rules: 1. **Qualitative label always wins.** Numbers are secondary. 2. **Scores below 60 hide the number.** Use the label only — never make someone feel rejected by a digit. 3. **Most positive accurate framing.** Same chart can be read 5 ways; pick the one that respects the person. 4. **Today's advice is specific, not vague.** "Reach out before 2pm" beats "Mercury retrograde." If your AI client surfaces a numeric score when `should_show_score: false`, it's violating the design contract. --- ## Examples — sample prompts that route through asktian These are the kinds of user prompts that an AI agent with asktian installed handles dramatically better: | User prompt | Tool the AI will use | |---|---| | *"How will today be for me? Born 1992-05-15."* | `asktian_daily_reading` | | *"I have a 3pm meeting tomorrow with a difficult client. Should I move it?"* | `asktian_best_time_for_action` | | *"My partner is born 1990-08-22 and I'm 1992-05-15. How are we as a couple?"* | `asktian_compatibility` | | *"What kind of day is it today?"* | `asktian_today_energy` | | *"Is 'Aurora' a good name for my startup?"* | `asktian_name_analysis` | | *"Best week to launch my product? My birthdate is..."* | `asktian_best_time_for_action` | | *"Should I propose to my partner this month? Mine 1992-05-15, theirs 1990-08-22."* | `asktian_best_time_for_action` + `asktian_compatibility` | The agent's response in each case becomes specific and actionable instead of vague. --- ## Roadmap - [ ] Direct integration with the full api.asktian.com endpoint catalogue (qimen, ziwei, almanac, fengshui, name-analysis traditional) - [ ] Lucky places near me (Google Places integration) - [ ] Tian-points / $TIAN token incentives for power use - [ ] Tools for venue / business: "is this address auspicious for a cafe" - [ ] Multi-tradition cross-reference (Western astrology, Vedic, I Ching) — same person, multiple lenses PRs welcome. The lib/ folder is intentionally self-contained for clean npm distribution. --- ## Links - **API & docs (optional key — higher limits + premium tier):** [api.asktian.com](https://api.asktian.com) - **$TIAN token + wallet:** [wallet.asktian.com](https://wallet.asktian.com) - **Source:** [github.com/douglasgan/asktian-mcp](https://github.com/douglasgan/asktian-mcp) --- ## License MIT — use it anywhere, fork it, ship it inside your own product. The asktian protocol is meant to be public infrastructure. 不知道就问天.