Let AI agents
understand
Taiwan healthcare.

Open-source MCP server for FHIR Box, TW Core IG and Taiwan terminology. Free for dev, hosted for prod.

63 tools · TW Core / TW NHI PAS · TFDA drugs · ICD-10 / SNOMED / LOINC / RxNorm · MIT · Passed Taiwan Connectathon 2026

63
MCP Tools
12
Tool Groups
7
Data Standards
01 / How it works

From MCP handshake to clinical answers in seconds.

Taiwan Health MCP follows the Model Context Protocol — the same standard used by Claude, ChatGPT, and dozens of agent frameworks. No custom SDK, no proprietary format.

01

Connect your LLM client

Add the MCP endpoint to Claude Desktop, ChatGPT, or any MCP-compatible agent. The handshake negotiates all 63 tools automatically.

02

Browse the tool catalog

Your AI assistant discovers every tool by name and schema — from ICD-10 code search to TFDA pill identification to FHIR R4 validation.

03

Query in natural language

Describe what you need: "Identify this pill based on its imprint" or "Find all LOINC codes for HbA1c." The LLM picks the right tool and returns structured results.

04

Scale freely

Deploy via Docker for on-premise use, or connect to the managed endpoint. MIT licensed — fork and customize freely.

02 / Tools

One MCP endpoint. Every Taiwan healthcare dataset your LLM needs.

Taiwan Health MCP unifies TFDA drug records, FHIR R4 profiles, ICD-10 / LOINC / SNOMED CT terminologies, and curated Taiwan clinical content behind the official Model Context Protocol — ready to plug into Claude, ChatGPT, or any agent runtime.

claude — taiwan-health-mcp
$claude mcp add taiwan-health http://localhost:8080/mcp
✓ taiwan-health connected
>identify_unknown_pill { shape: "oblong", color: "white", imprint: "BMS" }
Querying TFDA 仿單 + OCR index...
3 candidates · top match: Eliquis 5mg

Connect Taiwan Health MCP

Add Taiwan healthcare data to any MCP-compatible LLM client.

T
Claude Desktop Setup
claude mcp add taiwan-health https://tw-health-mcp.healthymind-tech.com/mcp
  1. 1Open Claude Desktop
  2. 2Add the MCP endpoint to your config
  3. 3Restart Claude — tools appear automatically
Want to self-host?Deploy Guide →
03 / Connect

Seamless MCP Integration for Clinical AI

Compatible with Claude Desktop, ChatGPT, and any MCP-enabled AI assistant.

01

Taiwan-Localized & International

Covers TFDA drug database (外觀/仿單/OCR), health supplements, food nutrition and clinical calculators — alongside ICD-10-CM/PCS (CMS 2026-04-01), SNOMED CT, LOINC 2.80 and RxNorm.

02

Production-Grade Infrastructure

PostgreSQL 16 with pgvector for semantic / hybrid search via Ollama embeddings, with Redis for real-time admin events and MinIO for pill images and leaflets.

03

Dual Interface: MCP + OpenAPI

Every tool is available natively through MCP (port /mcp) and as a RESTful OpenAPI endpoint at /tools/<name> for non-MCP integrations.

Every LLM Deserves Quality Clinical Data

Taiwan Health MCP treats AI agents as first-class citizens. Every tool is documented and typed — so any MCP-compatible LLM can discover, reason over, and combine Taiwan healthcare datasets autonomously.

Browse the Tool Catalog
04 / Open Source

Free. Open source. No lock-in.

No API keys. No paywalls. No rate limits. MIT licensed — self-host with Docker, or try the public endpoint first.

MIT Licensed

Full source on GitHub. Fork it, modify it, deploy it — no strings attached.

Self-Host Ready

Run the entire server on your own infrastructure with Docker. Your data stays local.

Community Driven

Open issues, discussions, and contributions welcome. No vendor lock-in, ever.

05 / Deployment

Three ways to run it. One set of tools.

Use our hosted endpoint, run it in your own cloud, or keep everything inside the hospital network.

Coming soon

Hosted (we run it)

We operate Taiwan Health MCP for you — data loading and updates, upgrades and an authenticated endpoint — so you only point your AI client at it.

Self-hosted (your cloud or data centre)

git clone, then docker compose up -d: one nginx gateway (:8080) serves both MCP and OpenAPI, with PostgreSQL 16 + pgvector, Redis and MinIO in the same compose file. MIT-licensed, no usage limits.

Offline, on-premises hospital

No cloud dependency: upload terminology source files and FHIR IG packages (package.tgz) through the admin console, run semantic search on a local Ollama model, and register the hospital's own FHIR Box.

Deployment guide ↗
06 / Auth & security

Authentication that matches the environment.

No tier stores patient data: audit records keep only the tool name and a SHA-256 hash of its parameters, never the raw values.

01

Community public endpoint / self-host default

No auth

Serves only public terminology and drug data, no PHI — suited to development, testing and trusted internal networks.

02

Self-hosted, publicly exposed

Bearer token

Set PUBLIC_TOOLS_AUTH_MODE=bearer and /mcp, /openapi.json and /tools/* all require Authorization: Bearer <token>; CORS origins may not include *.

03

Hosted and Enterprise

OAuth / SSO

Integrates with your organisation's identity provider, with access control and audit logs, for production and hospital deployments.

07 / Plans

Free for dev, hosted for prod.

The open-source Community edition stays free. Hosted and Enterprise are for teams that would rather not run it themselves.

Community

Free

Forever

  • MIT licence
  • All 63 tools
  • Self-host with no usage limits
  • Community support on GitHub Issues
View on GitHub

Hosted

Coming soon

Monthly subscription

  • Operated and updated by us
  • Authenticated production endpoint
  • Data versions kept current
  • No servers to run
Join the waitlist

Enterprise

Contact us

Quoted per deployment

  • OAuth / SSO integration
  • Audit logs and access control
  • On-premises hospital rollout
  • Technical support
Contact us

Open-source. Production-ready. Taiwan-first.

Taiwan Health MCP is built for healthcare developers, clinical informaticists, and AI application builders who need structured Taiwanese and international medical data via LLM-compatible tooling.

This service is provided for development and research purposes. Always validate AI-generated clinical outputs before deploying to production environments.

Star on GitHub