import glob
import os
from mcp.server.fastmcp import FastMCP
from config import MCPConfig # Import at top level
from clinical_guideline_service import ClinicalGuidelineService
from drug_service import DrugService
from fhir_condition_service import FHIRConditionService
from fhir_medication_service import FHIRMedicationService
from food_nutrition_service import FoodNutritionService
from health_food_service import HealthFoodService
from icd_service import ICDService
from lab_service import LabService
from utils import log_error, log_info
class DisabledService:
def __init__(self, name: str) -> None:
self.name = name
def __getattr__(self, _name: str):
def _disabled(*_args, **_kwargs):
return f"{self.name} service is disabled in this deployment."
return _disabled
def parse_enabled_modules() -> set[str] | None:
raw = os.getenv("MCP_MODULES", "").strip()
if not raw:
return None
return {item.strip().lower() for item in raw.split(",") if item.strip()}
enabled_modules = parse_enabled_modules()
def is_enabled(module: str) -> bool:
return enabled_modules is None or module.lower() in enabled_modules
def init_service(label: str, module: str, init_fn):
if not is_enabled(module):
log_info(f"{label} disabled by MCP_MODULES")
return DisabledService(label)
try:
return init_fn()
except Exception as e:
log_error(f"{label} failed: {e}")
return DisabledService(label)
# 0. Load Configuration
config = MCPConfig.from_env()
# 1. Initialize the MCP Server
# Pass host and port here, as run() does not accept them in newer versions
mcp = FastMCP(
"taiwanHealthMcp",
host=config.host,
port=config.port,
dependencies=["uvicorn"],
)
# 2. Configure data paths
# Automatically detect if running in Google Colab or Docker
if os.path.exists("/content/Taiwan-Health-MCP/data"):
DATA_DIR = "/content/Taiwan-Health-MCP/data"
elif os.path.exists("/app/data"):
DATA_DIR = "/app/data"
else:
# Fallback to local data directory
DATA_DIR = os.path.join(os.path.dirname(os.path.dirname(__file__)), "data")
log_info(f"Using DATA_DIR: {DATA_DIR}")
# Automatically find the ICD-10 Excel file.
excel_files = glob.glob(os.path.join(DATA_DIR, "*.xlsx"))
if excel_files:
ICD_FILE_PATH = excel_files[0]
log_info(f"Found ICD Excel file: {ICD_FILE_PATH}")
else:
ICD_FILE_PATH = os.path.join(DATA_DIR, "default.xlsx")
log_error("No Excel file found in data directory!")
# 3. Initialize Services with individual try-except blocks to ensure maximum availability
log_info("Initializing Services...")
icd_service = init_service("ICD", "icd", lambda: ICDService(ICD_FILE_PATH, DATA_DIR))
drug_service = init_service("Drug", "drug", lambda: DrugService(DATA_DIR))
health_food_service = init_service(
"Health Food", "health_food", lambda: HealthFoodService(DATA_DIR)
)
food_nutrition_service = init_service(
"Food Nutrition", "food_nutrition", lambda: FoodNutritionService(DATA_DIR)
)
lab_service = init_service("Lab", "lab", lambda: LabService(DATA_DIR))
guideline_service = init_service(
"Guideline", "guideline", lambda: ClinicalGuidelineService(DATA_DIR)
)
if is_enabled("fhir_condition") and not isinstance(icd_service, DisabledService):
fhir_condition_service = init_service(
"FHIR Condition", "fhir_condition", lambda: FHIRConditionService(icd_service)
)
else:
fhir_condition_service = DisabledService("FHIR Condition")
if is_enabled("fhir_medication") and not isinstance(drug_service, DisabledService):
fhir_medication_service = init_service(
"FHIR Medication",
"fhir_medication",
lambda: FHIRMedicationService(drug_service),
)
else:
fhir_medication_service = DisabledService("FHIR Medication")
# ==========================================
# Group 1: ICD-10 Tools (Diagnosis & Procedures)
# ==========================================
@mcp.tool()
def search_medical_codes(keyword: str, type: str = "all") -> str:
"""
Search for ICD-10-CM (Diagnosis) or ICD-10-PCS (Procedure) codes.
Args:
keyword: Search term (e.g., 'Diabetes', 'E11', 'Appendectomy', '子宮內膜異位').
type: Filter by 'diagnosis', 'procedure', or 'all'. Default is 'all'.
"""
log_info(f"Tool called: search_medical_codes with query='{keyword}', type='{type}'")
return icd_service.search_codes(keyword, type)
@mcp.tool()
def infer_complications(code: str) -> str:
"""
Infers potential complications or specific sub-conditions based on ICD hierarchy.
Example: Input 'E11' (Type 2 Diabetes) -> Returns specific codes like E11.9, E11.2 (with kidney complications).
Args:
code: The base diagnosis code (e.g., 'E11', 'N80').
"""
log_info(f"Tool called: infer_complications with code='{code}'")
return icd_service.infer_complications(code)
@mcp.tool()
def get_nearby_codes(code: str) -> str:
"""
Retrieves codes immediately preceding and following the target code.
Useful for differential diagnosis context or seeing related severity levels.
Args:
code: The target diagnosis code.
"""
log_info(f"Tool called: get_nearby_codes with code='{code}'")
return icd_service.get_nearby_codes(code)
@mcp.tool()
def check_medical_conflict(diagnosis_code: str, procedure_code: str) -> str:
"""
**IMPORTANT: Use this tool when you need to verify if a diagnosis and procedure combination is medically appropriate.**
This tool simultaneously retrieves and compares:
- Full details of the diagnosis code (ICD-10-CM)
- Full details of the procedure code (ICD-10-PCS)
- Provides structured data for conflict analysis
Use this when the user asks about:
- "Are these codes compatible?"
- "Does X diagnosis match Y procedure?"
- "Check if this procedure is appropriate for this diagnosis"
- "Is there a conflict between these codes?"
Args:
diagnosis_code: ICD-10-CM diagnosis code (e.g., 'K35.80', 'N80.A0').
procedure_code: ICD-10-PCS procedure code (e.g., '0DTJ0ZZ', '0UT90ZZ').
Returns:
JSON with both diagnosis and procedure details for comparison and conflict analysis.
"""
log_info(
f"Tool called: check_medical_conflict ({diagnosis_code} vs {procedure_code})"
)
return icd_service.get_conflict_info(diagnosis_code, procedure_code)
# ==========================================
# Group 2: Drug Tools (Taiwan FDA Data)
# ==========================================
@mcp.tool()
def search_drug_info(keyword: str) -> str:
"""
Search for Taiwan FDA approved drugs by name (Chinese/English) or indication.
Returns basic information including License ID, Name, and Indication.
Args:
keyword: Drug name or symptom (e.g., 'Panadol', '普拿疼', '頭痛').
"""
log_info(f"Tool called: search_drug_info with query='{keyword}'")
return drug_service.search_drug(keyword)
@mcp.tool()
def get_drug_details(license_id: str) -> str:
"""
Get comprehensive details for a specific drug license ID.
Includes: Ingredients, Usage, Appearance (shape/color), and Package Insert links.
Args:
license_id: The specific license ID found via search (e.g., '衛部藥製字第058498號').
"""
log_info(f"Tool called: get_drug_details for ID='{license_id}'")
return drug_service.get_details(license_id)
@mcp.tool()
def identify_unknown_pill(features: str) -> str:
"""
Identify a pill based on visual features using the appearance database.
Args:
features: Keywords describing the pill (e.g., 'white circle YP', 'oval pink').
Include shape, color, and markings if visible.
"""
log_info(f"Tool called: identify_unknown_pill with features='{features}'")
return drug_service.identify_pill(features)
# ==========================================
# Group 3: Composite Analysis (The "Doctor Brain")
# ==========================================
@mcp.tool()
def analyze_treatment_plan(diagnosis_keyword: str, drug_keyword: str) -> str:
"""
[Advanced] Analyze the correlation between a diagnosis and a drug.
Fetches data from both ICD-10 and FDA Drug databases to help evaluate treatment appropriateness.
Args:
diagnosis_keyword: The condition or disease name/code (e.g., 'Diabetes', 'E11').
drug_keyword: The medication name (e.g., 'Metformin').
"""
log_info(
f"Tool called: analyze_treatment with diagnosis='{diagnosis_keyword}', drug='{drug_keyword}'"
)
# 1. Search ICD (broad search first)
icd_result = icd_service.search_codes(diagnosis_keyword, type="diagnosis")
# 2. Search Drug
drug_result = drug_service.search_drug(drug_keyword)
# 3. Combine results for the LLM to process
return f"""
=== Comprehensive Analysis Context ===
[Context 1: Diagnosis Data (ICD-10)]
Query: {diagnosis_keyword}
Results:
{icd_result}
-------------------
[Context 2: Medication Data (Taiwan FDA)]
Query: {drug_keyword}
Results:
{drug_result}
-------------------
SYSTEM INSTRUCTION:
Based on the above retrieved data, please analyze:
1. Does the medication's indication match the diagnosis?
2. Are there any obvious contraindications based on the drug category?
3. Provide a brief summary for a healthcare professional.
"""
# ==========================================
# Group 4: Health Food Tools (Taiwan FDA)
# ==========================================
@mcp.tool()
def search_health_food(keyword: str) -> str:
"""
Search for Taiwan FDA approved health foods by name or health benefit.
Returns information including license number, category, health benefits, and claims.
Args:
keyword: Product name or health benefit (e.g., '靈芝', '調節血脂', '護肝').
"""
log_info(f"Tool called: search_health_food with query='{keyword}'")
return health_food_service.search_health_food(keyword)
@mcp.tool()
def get_health_food_details(license_number: str) -> str:
"""
Get comprehensive details for a specific health food by license number.
Includes functional components, health benefits, claims, warnings, and precautions.
Args:
license_number: The specific license number (e.g., '衛部健食字第A00123號').
"""
log_info(f"Tool called: get_health_food_details for license='{license_number}'")
return health_food_service.get_health_food_details(license_number)
# ==========================================
# Group 5: Nutrition & Dietary Management Tools
# ==========================================
@mcp.tool()
def search_food_nutrition(food_name: str, nutrient: str = None) -> str:
"""
Search for nutritional information of foods from Taiwan's food nutrition database.
Args:
food_name: Name of the food (e.g., '白米', '雞蛋', '蘋果', 'chicken breast').
nutrient: Optional - specific nutrient to filter (e.g., '蛋白質', '維生素C', '鈣').
"""
log_info(
f"Tool called: search_food_nutrition with food='{food_name}', nutrient='{nutrient}'"
)
return food_nutrition_service.search_nutrition(food_name, nutrient)
@mcp.tool()
def get_detailed_nutrition(food_name: str) -> str:
"""
Get comprehensive nutritional breakdown for a specific food item.
Returns all available nutrients organized by category (e.g., macronutrients, vitamins, minerals).
Args:
food_name: The specific food name (e.g., '糙米', '雞胸肉').
"""
log_info(f"Tool called: get_detailed_nutrition for food='{food_name}'")
return food_nutrition_service.get_detailed_nutrition(food_name)
@mcp.tool()
def search_food_ingredient(keyword: str) -> str:
"""
Search for food ingredients/materials in Taiwan's regulatory database.
Useful for checking if an ingredient is approved for food use.
Args:
keyword: Ingredient name in Chinese or English (e.g., '薑黃', 'turmeric', '人參').
"""
log_info(f"Tool called: search_food_ingredient with query='{keyword}'")
return food_nutrition_service.search_food_ingredient(keyword)
@mcp.tool()
def get_ingredients_by_category(category: str) -> str:
"""
Get all approved food ingredients in a specific category.
Args:
category: Category name (e.g., '香料植物', '食品添加物', '著色劑').
"""
log_info(f"Tool called: get_ingredients_by_category for category='{category}'")
return food_nutrition_service.get_ingredients_by_category(category)
@mcp.tool()
def analyze_meal_nutrition(foods: list[str]) -> str:
"""
Analyze the combined nutritional composition of multiple foods (meal planning).
Args:
foods: List of food names to analyze together (e.g., ['白米', '雞胸肉', '青花菜']).
"""
log_info(f"Tool called: analyze_meal_nutrition with foods={foods}")
return food_nutrition_service.analyze_diet_plan(foods)
# ==========================================
# Group 6: Comprehensive Health Analysis (疾病與保健整合分析)
# ==========================================
@mcp.tool()
def analyze_health_support_for_condition(diagnosis_keyword: str) -> str:
"""
**【綜合分析工具】疾病與健康食品輔助保健分析**
根據疾病診斷，提供適合的健康食品推薦與飲食建議。
⚠️ **重要法規聲明**：
- 健康食品非藥品，不具治療疾病之效能
- 此功能僅供「輔助保健參考」，不可取代正規醫療
- 所有建議必須在醫師指導下使用
**功能說明**：
1. 查詢疾病的 ICD-10 診斷資訊
2. 根據疾病對應適合的保健功效
3. 推薦台灣 FDA 核可的相關健康食品
4. 提供基於實證的飲食營養建議
5. 包含完整醫療免責聲明
**適用場景**：
- 慢性病患者尋求輔助保健建議
- 了解特定疾病可參考的健康食品
- 整合性健康管理規劃
Args:
diagnosis_keyword: 疾病名稱或 ICD-10 碼
Examples:
- ICD 碼: 'E11' (第二型糖尿病), 'I10' (高血壓), 'K74' (肝硬化)
- 中文: '糖尿病', '高血脂', '骨質疏鬆'
- 英文: 'diabetes', 'hypertension', 'osteoporosis'
Returns:
完整的綜合分析報告，包含：
- 疾病診斷資訊（來自 ICD-10）
- 建議的保健功效類別
- 相關健康食品列表（台灣 FDA 核可）
- 飲食營養建議
- 醫療安全聲明
**使用範例**：
- analyze_health_support_for_condition("E11") # 查詢第二型糖尿病
- analyze_health_support_for_condition("糖尿病") # 使用中文查詢
- analyze_health_support_for_condition("高血壓") # 查詢高血壓
"""
log_info(
f"Tool called: analyze_health_support_for_condition with diagnosis='{diagnosis_keyword}'"
)
# 傳入 icd_service 和 food_nutrition_service 以整合疾病資訊和飲食建議
return health_food_service.analyze_health_support_for_condition(
diagnosis_keyword=diagnosis_keyword,
icd_service=icd_service,
food_nutrition_service=food_nutrition_service,
)
# ==========================================
# Group 7: FHIR Interoperability Tools
# ==========================================
@mcp.tool()
def create_fhir_condition(
icd_code: str,
patient_id: str,
clinical_status: str = "active",
verification_status: str = "confirmed",
category: str = "encounter-diagnosis",
severity: str = None,
onset_date: str = None,
recorded_date: str = None,
additional_notes: str = None,
) -> str:
"""
將 ICD-10-CM 診斷碼轉換為符合 FHIR R4 標準的 Condition 資源。
FHIR (Fast Healthcare Interoperability Resources) 是全球醫療資訊交換標準，
此工具可生成符合台灣醫療系統的 FHIR Condition 資源。
Args:
icd_code: ICD-10-CM 診斷碼（例如: "E11.9" 表示第二型糖尿病）
patient_id: 患者識別碼（例如: "patient-001"）
clinical_status: 臨床狀態，可選值:
- "active": 活動中（症狀持續存在）
- "inactive": 非活動（無症狀但未完全治癒）
- "resolved": 已解決（完全治癒）
- "remission": 緩解期（癌症等疾病的緩解狀態）
verification_status: 驗證狀態，可選值:
- "confirmed": 已確診（經檢查確認）
- "provisional": 臨時診斷（初步診斷，需進一步確認）
- "differential": 鑑別診斷（多個可能診斷之一）
- "refuted": 已排除（排除此診斷）
category: 診斷分類，可選值:
- "encounter-diagnosis": 就診診斷（本次就醫的診斷）
- "problem-list-item": 問題清單項目（長期追蹤的健康問題）
severity: 嚴重程度（可選），可選值: "mild"（輕度）, "moderate"（中度）, "severe"（重度）
onset_date: 發病日期（可選），格式: YYYY-MM-DD（例如: "2024-01-15"）
recorded_date: 記錄日期時間（可選），格式: YYYY-MM-DDTHH:MM:SS+08:00
additional_notes: 額外備註（可選），任何補充說明
Returns:
符合 FHIR R4 標準的 Condition 資源（JSON 格式字串）
Example:
create_fhir_condition(
icd_code="E11.9",
patient_id="patient-001",
clinical_status="active",
severity="moderate",
onset_date="2024-01-15"
)
"""
log_info(
f"Tool called: create_fhir_condition for ICD={icd_code}, Patient={patient_id}"
)
result = fhir_condition_service.create_condition(
icd_code=icd_code,
patient_id=patient_id,
clinical_status=clinical_status,
verification_status=verification_status,
category=category,
severity=severity,
onset_date=onset_date,
recorded_date=recorded_date,
additional_notes=additional_notes,
)
return fhir_condition_service.to_json_string(result, indent=2)
@mcp.tool()
def create_fhir_condition_from_diagnosis(
diagnosis_keyword: str,
patient_id: str,
clinical_status: str = "active",
verification_status: str = "confirmed",
severity: str = None,
) -> str:
"""
從疾病關鍵字搜尋並建立 FHIR Condition 資源（自動查找 ICD-10 編碼）。
此工具會先搜尋符合關鍵字的診斷碼，然後自動建立 FHIR Condition 資源。
適合用於不確定 ICD-10 編碼但知道疾病名稱的情況。
Args:
diagnosis_keyword: 疾病關鍵字（中文或英文），例如: "糖尿病", "高血壓", "Diabetes"
patient_id: 患者識別碼
clinical_status: 臨床狀態（預設: "active"）
verification_status: 驗證狀態（預設: "confirmed"）
severity: 嚴重程度（可選）
Returns:
包含搜尋結果和 FHIR Condition 資源的 JSON 字串
Example:
create_fhir_condition_from_diagnosis(
diagnosis_keyword="第二型糖尿病",
patient_id="patient-001",
severity="moderate"
)
"""
log_info(
f"Tool called: create_fhir_condition_from_diagnosis for '{diagnosis_keyword}'"
)
result = fhir_condition_service.create_condition_from_search(
keyword=diagnosis_keyword,
patient_id=patient_id,
clinical_status=clinical_status,
verification_status=verification_status,
severity=severity,
)
return fhir_condition_service.to_json_string(result, indent=2)
@mcp.tool()
def validate_fhir_condition(condition_json: str) -> str:
"""
驗證 FHIR Condition 資源是否符合 FHIR R4 標準規範。
此工具會檢查 Condition 資源的必要欄位、資料格式和邏輯正確性。
Args:
condition_json: FHIR Condition 資源的 JSON 字串
Returns:
驗證結果，包含錯誤和警告訊息
Example:
validate_fhir_condition('{"resourceType": "Condition", ...}')
"""
log_info("Tool called: validate_fhir_condition")
import json
try:
condition = json.loads(condition_json)
result = fhir_condition_service.validate_condition(condition)
return fhir_condition_service.to_json_string(result, indent=2)
except json.JSONDecodeError as e:
log_error(f"JSON decode error in validate_fhir_condition: {e}")
return fhir_condition_service.to_json_string(
{"valid": False, "errors": [f"Invalid JSON format: {str(e)}"]}, indent=2
)
# ==========================================
# Group 8: Laboratory & LOINC Tools
# ==========================================
@mcp.tool()
def search_loinc_code(keyword: str, category: str = None) -> str:
"""
搜尋 LOINC 碼（台灣檢驗項目對照國際標準）
LOINC (Logical Observation Identifiers Names and Codes) 是全球醫療檢驗項目的標準編碼系統。
本工具提供台灣常用檢驗項目與 LOINC 碼的對照查詢。
Args:
keyword: 搜尋關鍵字（檢驗名稱、LOINC 碼、常用縮寫）
- 支援中文：例如 "血糖"、"肝功能"、"糖化血色素"
- 支援英文：例如 "Glucose"、"HbA1c"
- 支援縮寫：例如 "WBC"、"RBC"、"ALT"、"Cr"
category: 分類篩選（可選）
- "血液常規"、"生化檢驗-血糖"、"生化檢驗-血脂"
- "生化檢驗-肝功能"、"生化檢驗-腎功能"
- "內分泌-甲狀腺"、"凝血功能"、"發炎指標"
Returns:
LOINC 碼、檢驗名稱（中英文）、檢體類型、單位、檢測方法
Examples:
- search_loinc_code("血糖") # 查詢血糖相關檢驗
- search_loinc_code("HbA1c") # 查詢糖化血色素
- search_loinc_code("肝功能", category="生化檢驗-肝功能")
"""
log_info(
f"Tool called: search_loinc_code with keyword='{keyword}', category='{category}'"
)
return lab_service.search_loinc_code(keyword, category)
@mcp.tool()
def list_lab_categories() -> str:
"""
列出所有檢驗分類
Returns:
所有可用的檢驗分類清單
"""
log_info("Tool called: list_lab_categories")
return lab_service.list_categories()
@mcp.tool()
def get_reference_range(loinc_code: str, age: int, gender: str = "all") -> str:
"""
查詢檢驗參考值範圍（依年齡、性別）
每個檢驗項目都有正常參考值範圍，且會因年齡、性別而異。
本工具提供台灣醫療機構常用的參考值標準。
Args:
loinc_code: LOINC 碼（可使用 search_loinc_code 工具查詢）
age: 患者年齡（歲）
gender: 性別
- "M": 男性
- "F": 女性
- "all": 不分性別（預設）
Returns:
檢驗項目資訊、參考值範圍（上限/下限）、適用對象
Examples:
- get_reference_range("1558-6", age=45, gender="M") # 45歲男性空腹血糖參考值
- get_reference_range("718-7", age=30, gender="F") # 30歲女性血紅素參考值
"""
log_info(
f"Tool called: get_reference_range for LOINC={loinc_code}, age={age}, gender={gender}"
)
return lab_service.get_reference_range(loinc_code, age, gender)
@mcp.tool()
def interpret_lab_result(
loinc_code: str, value: float, age: int, gender: str = "all"
) -> str:
"""
判讀檢驗結果（自動比對參考值，判斷是否異常）
輸入檢驗數值，系統會自動與參考值比對，判斷是否偏高、偏低或正常。
Args:
loinc_code: LOINC 碼
value: 檢驗數值
age: 患者年齡（歲）
gender: 性別（"M"=男性, "F"=女性, "all"=不分性別）
Returns:
檢驗結果判讀：
- 數值狀態（正常/偏高/偏低）
- 參考值範圍
- 臨床意義
Examples:
- interpret_lab_result("1558-6", value=126, age=50, gender="M")
# 判讀空腹血糖 126 mg/dL
- interpret_lab_result("4548-4", value=7.5, age=60, gender="F")
# 判讀 HbA1c 7.5%
"""
log_info(f"Tool called: interpret_lab_result for LOINC={loinc_code}, value={value}")
return lab_service.interpret_lab_result(loinc_code, value, age, gender)
@mcp.tool()
def batch_interpret_lab_results(
results_json: str, age: int, gender: str = "all"
) -> str:
"""
批次判讀多個檢驗結果
一次判讀多個檢驗項目，快速找出異常項目。
Args:
results_json: 檢驗結果 JSON 字串
格式: [{"loinc_code": "1558-6", "value": 126}, ...]
age: 患者年齡
gender: 性別
Returns:
批次判讀結果摘要、異常項目統計、各項目詳細判讀
Example:
batch_interpret_lab_results(
results_json='[
{"loinc_code": "1558-6", "value": 126},
{"loinc_code": "4548-4", "value": 7.2},
{"loinc_code": "2093-3", "value": 220}
]',
age=55,
gender="M"
)
"""
log_info(f"Tool called: batch_interpret_lab_results for age={age}, gender={gender}")
import json
try:
results = json.loads(results_json)
return lab_service.batch_interpret_results(results, age, gender)
except json.JSONDecodeError as e:
log_error(f"JSON decode error in batch_interpret_lab_results: {e}")
return json.dumps(
{"error": f"Invalid JSON format: {str(e)}"}, ensure_ascii=False
)
# ==========================================
# Group 9: Clinical Guideline Tools
# ==========================================
@mcp.tool()
def search_clinical_guideline(keyword: str) -> str:
"""
搜尋臨床診療指引
提供台灣各醫學會發布的臨床診療指引，包含診斷、治療、用藥建議。
Args:
keyword: 疾病名稱或 ICD-10 編碼
- 例如："糖尿病"、"高血壓"、"E11"、"I10"
Returns:
診療指引清單（標題、來源醫學會、發布年份）
Examples:
- search_clinical_guideline("糖尿病")
- search_clinical_guideline("E11")
"""
log_info(f"Tool called: search_clinical_guideline with keyword='{keyword}'")
return guideline_service.search_guideline(keyword)
@mcp.tool()
def get_complete_guideline(icd_code: str) -> str:
"""
取得完整診療指引（診斷、用藥、檢查、治療目標）
提供特定疾病的完整診療指引，包含：
- 診斷建議（檢查流程、診斷標準）
- 用藥建議（第一線/第二線藥物、劑量、禁忌症）
- 檢查建議（追蹤項目、頻率）
- 治療目標（血糖/血壓/血脂等控制目標）
Args:
icd_code: ICD-10 編碼（例如: "E11" 糖尿病, "I10" 高血壓）
Returns:
完整診療指引（結構化資料）
Examples:
- get_complete_guideline("E11") # 糖尿病完整指引
- get_complete_guideline("I10") # 高血壓完整指引
"""
log_info(f"Tool called: get_complete_guideline for ICD={icd_code}")
return guideline_service.get_complete_guideline(icd_code)
@mcp.tool()
def get_medication_recommendations(icd_code: str) -> str:
"""
取得用藥建議（根據診療指引）
提供特定疾病的標準用藥建議，包含第一線、第二線藥物選擇。
Args:
icd_code: ICD-10 編碼
Returns:
用藥建議清單（藥物分類、範例藥品、劑量指引、禁忌症、證據等級）
Example:
get_medication_recommendations("E11") # 糖尿病用藥建議
"""
log_info(f"Tool called: get_medication_recommendations for ICD={icd_code}")
return guideline_service.get_medication_recommendations(icd_code)
@mcp.tool()
def get_test_recommendations(icd_code: str) -> str:
"""
取得檢查建議（根據診療指引）
提供特定疾病的檢查建議，包含檢查項目、頻率、適應症。
Args:
icd_code: ICD-10 編碼
Returns:
檢查建議清單（檢查名稱、LOINC 碼、頻率、適應症、證據等級）
Example:
get_test_recommendations("E11") # 糖尿病檢查建議
"""
log_info(f"Tool called: get_test_recommendations for ICD={icd_code}")
return guideline_service.get_test_recommendations(icd_code)
@mcp.tool()
def get_treatment_goals(icd_code: str) -> str:
"""
取得治療目標（根據診療指引）
提供特定疾病的治療目標，例如血糖/血壓/血脂的控制目標值。
Args:
icd_code: ICD-10 編碼
Returns:
治療目標清單（目標類型、參數、目標值、時間範圍）
Example:
get_treatment_goals("E11") # 糖尿病治療目標（HbA1c <7% 等）
"""
log_info(f"Tool called: get_treatment_goals for ICD={icd_code}")
return guideline_service.get_treatment_goals(icd_code)
@mcp.tool()
def suggest_clinical_pathway(icd_code: str, patient_context_json: str = None) -> str:
"""
建議臨床路徑（完整治療流程）
根據診療指引，提供從診斷到治療的完整臨床路徑建議。
Args:
icd_code: ICD-10 編碼
patient_context_json: 患者背景資訊（可選）
格式: {"age": 60, "gender": "M", "comorbidities": ["高血壓", "高血脂"]}
Returns:
結構化臨床路徑：
- 步驟1: 診斷確認
- 步驟2: 基礎檢查
- 步驟3: 治療啟始
- 步驟4: 追蹤監測
- 步驟5: 治療目標
Example:
suggest_clinical_pathway("E11") # 糖尿病臨床路徑
"""
log_info(f"Tool called: suggest_clinical_pathway for ICD={icd_code}")
import json
patient_context = None
if patient_context_json:
try:
patient_context = json.loads(patient_context_json)
except json.JSONDecodeError as e:
log_error(f"JSON decode error in suggest_clinical_pathway: {e}")
pass
return guideline_service.suggest_clinical_pathway(icd_code, patient_context)
# ==========================================
# Group 10: FHIR Medication Tools
# ==========================================
@mcp.tool()
def create_fhir_medication(
license_id: str, include_ingredients: bool = True, include_appearance: bool = True
) -> str:
"""
將台灣 FDA 藥品許可證資料轉換為符合 FHIR R4 標準的 Medication 資源。
FHIR Medication 資源用於描述藥品的基本資訊，包含：
- 藥品識別資訊（許可證字號、藥品名稱）
- 製造商資訊
- 劑型（錠劑、膠囊、注射劑等）
- 成分資訊（有效成分與含量）
- 外觀描述（形狀、顏色、刻痕）
Args:
license_id: 台灣 FDA 藥品許可證字號
範例: "衛署藥製字第012345號", "衛部藥輸字第026123號"
include_ingredients: 是否包含成分資訊（預設: True）
include_appearance: 是否包含外觀描述（預設: True）
Returns:
符合 FHIR R4 標準的 Medication 資源（JSON 格式字串）
Example:
create_fhir_medication(
license_id="衛署藥製字第058498號",
include_ingredients=True,
include_appearance=True
)
"""
log_info(f"Tool called: create_fhir_medication for license_id={license_id}")
result = fhir_medication_service.create_medication(
license_id=license_id,
include_ingredients=include_ingredients,
include_appearance=include_appearance,
)
return fhir_medication_service.to_json_string(result, indent=2)
@mcp.tool()
def create_fhir_medication_knowledge(license_id: str) -> str:
"""
建立 FHIR MedicationKnowledge 資源（藥品知識庫）。
MedicationKnowledge 提供比 Medication 更詳細的藥品知識，包含：
- 適應症（Indication）
- 用法用量（Dosage Guidelines）
- ATC 藥物分類（Anatomical Therapeutic Chemical Classification）
- 包裝資訊
- 藥品類別（處方藥/非處方藥）
- 藥品特性（顏色、形狀、刻痕、外觀圖片）
適用於：
- 藥品資訊系統整合
- 臨床決策支援系統
- 藥品知識庫建立
- 與國際醫療系統對接
Args:
license_id: 台灣 FDA 藥品許可證字號
Returns:
符合 FHIR R4 標準的 MedicationKnowledge 資源（JSON 格式字串）
Example:
create_fhir_medication_knowledge("衛署藥製字第058498號")
"""
log_info(
f"Tool called: create_fhir_medication_knowledge for license_id={license_id}"
)
result = fhir_medication_service.create_medication_knowledge(license_id)
return fhir_medication_service.to_json_string(result, indent=2)
@mcp.tool()
def create_fhir_medication_from_name(
drug_name: str, resource_type: str = "Medication"
) -> str:
"""
從藥品名稱搜尋並建立 FHIR 資源（自動查找許可證）。
此工具會先搜尋符合名稱的藥品，然後自動建立 FHIR 資源。
適合用於不確定許可證字號但知道藥品名稱的情況。
Args:
drug_name: 藥品名稱（中文或英文）
範例: "普拿疼", "Panadol", "立普妥"
resource_type: FHIR 資源類型
- "Medication": 基本藥品資源（預設）
- "MedicationKnowledge": 完整藥品知識庫
Returns:
包含搜尋結果和 FHIR 資源的 JSON 字串
Example:
create_fhir_medication_from_name(
drug_name="普拿疼",
resource_type="Medication"
)
"""
log_info(
f"Tool called: create_fhir_medication_from_name for drug_name='{drug_name}'"
)
result = fhir_medication_service.create_medication_from_search(
keyword=drug_name, resource_type=resource_type
)
return fhir_medication_service.to_json_string(result, indent=2)
@mcp.tool()
def identify_pill_to_fhir(
shape: str = None, color: str = None, marking: str = None
) -> str:
"""
從藥品外觀識別並建立 FHIR Medication 資源。
利用藥品外觀特徵（形狀、顏色、刻痕）識別藥品，並轉換為 FHIR 資源。
適用於：
- 未知藥品的識別
- 藥品查驗登記
- 用藥安全確認
Args:
shape: 形狀（可選）
範例: "圓形", "橢圓形", "菱形", "長方形", "circle", "oval"
color: 顏色（可選）
範例: "白色", "粉紅色", "藍色", "white", "pink", "blue"
marking: 刻痕/標記（可選）
範例: "YP", "500", "分割線"
Returns:
包含識別結果和 FHIR Medication 資源的 JSON 字串
Example:
identify_pill_to_fhir(
shape="圓形",
color="白色",
marking="500"
)
"""
log_info(
f"Tool called: identify_pill_to_fhir with shape={shape}, color={color}, marking={marking}"
)
result = fhir_medication_service.create_medication_from_appearance(
shape=shape, color=color, marking=marking
)
return fhir_medication_service.to_json_string(result, indent=2)
# --- Start Server ---
if __name__ == "__main__":
from config import MCPConfig
# 載入配置
config = MCPConfig.from_env()
# 輸出啟動資訊
log_info("=" * 50)
log_info("Taiwan Health MCP Server")
log_info("=" * 50)
log_info(str(config))
log_info("Server is starting...")
# 啟動服務
mcp.run(**config.get_run_kwargs())
