OpenTaiMed

Open-source tools for trustworthy Taiwan drug-information lookup — built on the public APIs of 衛福部食藥署 (Taiwan FDA / TFDA), made for AI agents.

English · 繁體中文 ↓

Important

OpenTaiMed is NOT an official TFDA product. It is an independent open-source wrapper around public TFDA APIs. The TFDA does not endorse or maintain it. Always check the official source before any clinical decision. This is not a medical device.

The idea

AI agents often answer drug questions by guessing from training data — and a brand name in one country can be a completely different drug in Taiwan. OpenTaiMed makes the assistant look up the real TFDA insert and cite it, instead of guessing.

LLM is a parser, not an author. Every answer must trace back to official TFDA text. If no source is found, it says "未載明" (not documented) — it never guesses.

What's in this repo

Part	Status	What it is
taiwan-fda-mcp/	shipped — v0.2.1	An MCP server with 3 tools. Works with any MCP client (Claude Desktop, Claude Code, Codex…).
Web backend	planned	API for non-MCP apps.
Web frontend	planned	A query page for clinicians.

Today taiwan-fda-mcp is the only shipped part — start there.

Try it

Once it's added to your AI agent, just ask a drug question in Chinese. The easiest install:

uvx taiwan-fda-mcp

Full setup + client configs (Claude Desktop / Claude Code / Codex) → taiwan-fda-mcp/README.md.

Where the data comes from

• mcp.fda.gov.tw — official package-insert (仿單) text, live.
• data.fda.gov.tw — license metadata, ingredients, etc. (cached daily).

Both are public TFDA APIs. No login, no scraping.

For developers

git clone https://github.com/shin13/opentaimed.git
cd opentaimed/taiwan-fda-mcp
uv sync && uv run pytest

Python 3.13 · uv · Ruff · Pyright · Conventional Commits. Report security issues privately — see SECURITY.md, not public issues.

License

MIT, with a clinical-safety disclaimer in the license file.

---

繁體中文

可信賴的台灣藥物資訊查詢開源工具,基於衛福部食藥署 (TFDA) 的公開 API,為 AI 代理人打造。

Important

OpenTaiMed 不是食藥署官方產品。 這是獨立的開源工具,只包裝 TFDA 的公開 API。 食藥署不背書、不維護本專案。臨床決策前請務必查核官方來源。本工具不是醫療器材。

核心理念

AI 代理人常用訓練資料「猜」藥物資訊——而某個國外商品名,在台灣可能是完全不同的藥。 OpenTaiMed 讓助理去查真正的 TFDA 仿單並附上出處,而不是用猜的。

LLM 是解析器,不是作者。 每個回答都必須回溯到 TFDA 官方原文;查無資料就說「未載明」,絕不臆測。

這個 repo 有什麼

部分	狀態	是什麼
taiwan-fda-mcp/	已上線 — v0.2.1	有 3 個工具的 MCP server,可搭配任何 MCP client(Claude Desktop、Claude Code、Codex…)。
後端 Web API	規劃中	給非 MCP 的應用使用。
前端網頁	規劃中	給臨床人員的查詢介面。

目前只有 taiwan-fda-mcp 上線,從這裡開始。

試用

把它加進你的 AI 代理人後,直接用中文問藥物問題。最簡單的安裝:

uvx taiwan-fda-mcp

完整安裝與 client 設定 → taiwan-fda-mcp/README.md。

資料來源

• mcp.fda.gov.tw — 官方仿單全文,即時。
• data.fda.gov.tw — 許可證 metadata、成分等(每日快取)。

兩者都是 TFDA 公開 API,免登入、不爬網頁。

給開發者

git clone https://github.com/shin13/opentaimed.git
cd opentaimed/taiwan-fda-mcp
uv sync && uv run pytest

Python 3.13 · uv · Ruff · Pyright · Conventional Commits。回報安全問題請私下進行—— 見 SECURITY.md,勿開公開 issue。

授權

MIT,授權檔內含臨床安全免責聲明。