unieai-mcp-accton-rfp 0.0.11__py3-none-any.whl

unieai_mcp_accton_rfp/server_v6.py

@@ -0,0 +1,349 @@
+import asyncio
+import json
+import logging
+import os
+import re
+import tempfile
+from typing import Any, Dict, Tuple, List, Optional
+from datetime import datetime
+
+import requests
+from dotenv import load_dotenv
+from fastmcp import FastMCP
+from openpyxl import load_workbook
+from openpyxl.worksheet.worksheet import Worksheet
+
+# LangChain (1.x API)
+from langchain_openai import ChatOpenAI
+from langchain_core.messages import HumanMessage, SystemMessage
+
+# ==============================
+# 🎛 Environment & Logging
+# ==============================
+
+load_dotenv()
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger("ExcelProcessor")
+
+app = FastMCP("ExcelProcessor")
+semaphore = asyncio.Semaphore(10)
+
+# LLM 初始化 (LangChain 1.x)
+llm = ChatOpenAI(
+    model=os.getenv("UNIEAI_MODEL"),
+    base_url=os.getenv("UNIEAI_API_URL"),
+    api_key=os.getenv("UNIEAI_API_KEY"),
+    temperature=0,
+    max_tokens=32768
+)
+
+# Appwrite ENV
+APPWRITE_PROJECT_ID = os.getenv("APPWRITE_PROJECT_ID")
+APPWRITE_API_KEY = os.getenv("APPWRITE_API_KEY")
+APPWRITE_ENDPOINT = os.getenv("APPWRITE_ENDPOINT", "https://sgp.cloud.appwrite.io/v1")
+
+
+# ==============================
+# 🧩 Helper Functions
+# ==============================
+
+def _extract_json(text: str) -> Dict[str, Any]:
+    """擷取 JSON 區塊"""
+    match = re.search(r"\{[\s\S]*\}", text)
+    if match:
+        try:
+            return json.loads(match.group(0))
+        except Exception as e:
+            logger.warning(f"JSON 解析失敗: {e}")
+    return {"Result": "解析錯誤", "Reference": text.strip()}
+
+
+def _parse_appwrite_url(url: str) -> Tuple[Optional[str], Optional[str]]:
+    pattern = r"/storage/buckets/([^/]+)/files/([^/]+)"
+    m = re.search(pattern, url)
+    if not m:
+        return None, None
+    return m.group(1), m.group(2)
+
+
+def _generate_new_filename(original_name: str) -> str:
+    base, ext = os.path.splitext(original_name)
+    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    return f"{base}_processed_{timestamp}{ext}"
+
+
+# ==============================
+# 🤖 LLM Logic（新增兩階段）
+# ==============================
+
+async def _call_llm_raw(prompt: str, user_message: str):
+    """返回 LLM 純文字內容"""
+    logger.info(f"🟢 _call_llm_raw : {prompt}, {user_message}")
+    try:
+        async with semaphore:
+            response = await llm.ainvoke([
+                SystemMessage(content=prompt),
+                HumanMessage(content=user_message)
+            ])
+            return (response.content or "").strip()
+    except Exception as e:
+        return f"LLM Error: {e}"
+
+
+def _extract_result_json(text: str):
+    """解析第二階段 JSON"""
+    try:
+        return json.loads(re.search(r"\{[\s\S]*\}", text).group(0))
+    except:
+        return {"Result": "Error"}
+
+
+# ==============================
+# 📘 Prompt 建構（新：兩個 prompt）
+# ==============================
+
+def _build_reference_prompt() -> str:
+    return """
+    你是一位嚴謹的產品經理助理，專門負責將內部產品規格（知識庫）與客戶的需求單（RFP）進行比對和符合性分析。
+
+    **任務指示：**
+    1.  你將收到客戶的產品需求單 (RFP) 作為輸入。
+    2.  你的知識庫已包含你公司產品的完整說明文件。
+    3.  請仔細閱讀 RFP 中的每一條具體需求，並利用你的產品知識庫內容進行嚴格比對。
+
+    **比對規則：**
+    * **Conform (完全符合)：** 公司的產品規格能**完整且無條件地**滿足 RFP 中的該項需求。
+    * **Half Conform (部分符合)：** 公司的產品規格**只能滿足** RFP 中該項需求的**部分內容**，或者需要透過**變通、額外配置或未來規劃**才能滿足。
+    * **Not Conform (不符合)：** 公司的產品規格**無法滿足** RFP 中的該項需求。
+
+    **輸出格式要求：**
+    你必須以條列式清晰地輸出分析結果，**每一條結果必須包含**：
+    1.  RFP 中的**原始需求描述** (簡短摘錄或編號)。
+    2.  **符合程度** (只能是：Conform, Half Conform, Not Conform 三者之一)。
+    3.  **參考依據** (說明做出判斷的依據，需明確引用知識庫中**相關產品說明**的關鍵資訊或段落，例如：知識庫中「功能A」的描述支持此判斷)。
+
+    請針對 RFP 中的每一條主要需求逐一進行分析。
+"""
+
+
+def _build_result_prompt() -> str:
+    return """
+請依據以下 Reference 文本，判斷其符合性：
+- Conform：完全符合
+- Half Conform：部分符合
+- Not Conform：不符合
+
+請僅輸出以下 JSON 格式：
+{
+  "Result": "Conform / Half Conform / Not Conform"
+}
+"""
+
+
+def _build_user_message(a: str, b: str, c: str, d: str) -> str:
+    logger.info(f"🟢 _build_user_message : {a}, {b}, {c}, {d}") 
+    return f"""
+{a}, {b}, {c}, {d}
+"""
+
+
+# ==============================
+# 📊 Excel Processing Core
+# ==============================
+
+async def _process_excel_logic(url: str) -> Dict[str, Any]:
+    logger.info(f"🟢 開始處理 Excel：{url}")
+
+    # -------------------------
+    # Step 1: Download / Load
+    # -------------------------
+    source_type = ""
+    local_path = None
+    appwrite_info = (None, None)
+    bucket_id = None
+
+    if url.startswith("file:///"):
+        local_path = url.replace("file:///", "")
+        file_path = local_path
+        source_type = "local"
+
+    elif url.startswith("http"):
+        resp = requests.get(url)
+        resp.raise_for_status()
+        with tempfile.NamedTemporaryFile(delete=False, suffix=".xlsx") as tmp:
+            tmp.write(resp.content)
+            file_path = tmp.name
+
+        bucket_id, file_id = _parse_appwrite_url(url)
+        if bucket_id:
+            source_type = "appwrite"
+            appwrite_info = (bucket_id, file_id)
+        else:
+            source_type = "remote_readonly"
+
+    else:
+        raise ValueError("❌ 不支援檔案來源")
+
+    # -------------------------
+    # Step 2: Open Excel
+    # -------------------------
+    wb = load_workbook(file_path)
+    ws = wb.active
+
+    header = {cell.value: idx for idx, cell in enumerate(ws[1], 1)}
+    for col in ["itemA", "itemB", "itemC", "itemD", "Result", "Reference"]:
+        if col not in header:
+            raise ValueError(f"❌ Excel 缺少欄位：{col}")
+
+    # -------------------------
+    # Step 3: Two-stage LLM (High Performance Version)
+    # -------------------------
+    rows_for_llm = []
+
+    for row in ws.iter_rows(min_row=2, values_only=False):
+        if any([cell.value for cell in row]):
+            rows_for_llm.append(row)
+
+    # Build input list for batch LLM calls
+    user_messages = []
+    for row in rows_for_llm:
+        a = row[header["itemA"] - 1].value or ""
+        b = row[header["itemB"] - 1].value or ""
+        c = row[header["itemC"] - 1].value or ""
+        d = row[header["itemD"] - 1].value or ""
+
+        user_messages.append(_build_user_message(str(a), str(b), str(c), str(d)))
+
+    reference_prompt = _build_reference_prompt()
+
+    # -------- Stage 1: Run all Reference LLM calls in parallel --------
+    ref_tasks = [
+        _call_llm_raw(reference_prompt, msg)
+        for msg in user_messages
+    ]
+
+    reference_results = await asyncio.gather(*ref_tasks)
+
+    # Write Reference to Excel
+    for row, ref_text in zip(rows_for_llm, reference_results):
+        r = row[0].row
+        ws.cell(r, header["Reference"], ref_text)
+
+    # -------- Stage 2: Run all Result LLM calls in parallel --------
+    result_prompt = _build_result_prompt()
+
+    result_tasks = [
+        _call_llm_raw(result_prompt, ref_text)
+        for ref_text in reference_results
+    ]
+
+    raw_result_outputs = await asyncio.gather(*result_tasks)
+
+    # Parse JSON + write Result to Excel
+    for row, raw_result in zip(rows_for_llm, raw_result_outputs):
+        r = row[0].row
+        parsed = _extract_result_json(raw_result)
+        ws.cell(r, header["Result"], parsed.get("Result", "Error"))
+
+
+    # -------------------------
+    # Step 4: Save local debug copy
+    # -------------------------
+    local_debug_dir = r"D:\TempExcelDebug"
+    os.makedirs(local_debug_dir, exist_ok=True)
+
+    local_debug_filename = _generate_new_filename("debug_output.xlsx")
+    local_debug_path = os.path.join(local_debug_dir, local_debug_filename)
+
+    wb.save(local_debug_path)
+    logger.info(f"📝 本機 debug 檔案已輸出：{local_debug_path}")
+
+    # -------------------------
+    # Step 5: Write back according to source
+    # -------------------------
+
+    # local
+    if source_type == "local":
+        wb.save(local_path)
+        return {
+            "status": "success",
+            "location_type": "local",
+            "output_path": local_path
+        }
+
+    # Appwrite
+    if source_type == "appwrite":
+        bucket_id, _ = appwrite_info
+
+        tmp_out_path = os.path.join(
+            tempfile.gettempdir(),
+            _generate_new_filename("upload.xlsx")
+        )
+        wb.save(tmp_out_path)
+
+        new_file_id = f"processed_{datetime.now().strftime('%Y%m%d_%H%M%S')}"
+        new_file_name = f"{new_file_id}.xlsx"
+
+        upload_url = f"{APPWRITE_ENDPOINT}/storage/buckets/{bucket_id}/files"
+
+        headers = {
+            "X-Appwrite-Project": APPWRITE_PROJECT_ID,
+            "X-Appwrite-Key": APPWRITE_API_KEY,
+        }
+
+        files = {
+            "file": (
+                new_file_name,
+                open(tmp_out_path, "rb"),
+                "application/vnd.openxmlformats-officedocument.spreadsheetml.sheet"
+            )
+        }
+
+        data = { "fileId": new_file_id }
+
+        resp = requests.post(upload_url, headers=headers, files=files, data=data)
+        resp.raise_for_status()
+
+        return {
+            "status": "success",
+            "location_type": "appwrite_new_file",
+            "file_id": new_file_id,
+            "file_name": new_file_name,
+            "upload_response": resp.json(),
+            "download_url": f"{APPWRITE_ENDPOINT}/storage/buckets/{bucket_id}/files/{new_file_id}/view?project={APPWRITE_PROJECT_ID}"
+        }
+
+    # remote (can't write back)
+    with tempfile.NamedTemporaryFile(delete=False, suffix=".xlsx") as tmp_out:
+        wb.save(tmp_out.name)
+        fallback = tmp_out.name
+
+    return {
+        "status": "success",
+        "location_type": "remote_readonly",
+        "output_path": fallback,
+        "message": "無法寫回遠端，只能輸出本機暫存檔"
+    }
+
+
+# ==============================
+# 🔧 MCP Tool
+# ==============================
+
+@app.tool()
+async def process_excel(url: str):
+    return await _process_excel_logic(url)
+
+
+# ==============================
+# 🚀 CLI Test
+# ==============================
+
+if __name__ == "__main__":
+    test_url = (
+        "https://sgp.cloud.appwrite.io/v1/storage/buckets/6904374b00056677a970/files/6937a7fb00180f83ab67/view?project=6901b22e0036150b66d3&mode=admin"
+    )
+    print("🚀 測試開始...")
+    result = asyncio.run(_process_excel_logic(test_url))
+    print(json.dumps(result, ensure_ascii=False, indent=2))