本專部分程式碼與文件由 AI 產生，我負責整體架構、除錯、測試與強化。

Bing Search Agent

Local Bing search service that fetches search results via Playwright and summarizes answers using local LLM. Supports both MCP Server and HTTP API access methods.

Requirements

• Python 3.10+
• Microsoft Edge or Google Chrome
• Local LLM service (LM Studio, Ollama, etc., compatible with OpenAI API format)

Installation

pip install fastapi uvicorn playwright beautifulsoup4 openai python-dotenv pydantic "mcp==1.6.0" "starlette>=0.40.0,<0.48.0"
playwright install

mcp >= 1.7 will conflict with fastapi's starlette version, please use mcp==1.6.0.

Configuration

Copy .env.example to .env and fill in the settings:

LLM_BASE_URL=http://localhost:1234/v1
LLM_API_KEY=lm-studio
LLM_MODEL=your-model-name

BROWSER_CHANNEL=msedge   # msedge or chrome
HEADLESS=false

HOST=127.0.0.1
PORT=61500

Starting

MCP Mode (Claude Desktop, Cursor, Windsurf)

python main.py mcp

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "bing-search": {
      "command": "python",
      "args": ["C:/path/to/main.py", "mcp"]
    }
  }
}

HTTP Mode (LangChain, AutoGen, OpenClaw)

python main.py http

After starting:

• API Docs: http://127.0.0.1:61500/docs
• Tool Manifest: http://127.0.0.1:61500/.well-known/ai-plugin.json

API

POST /search — Search and summarize

{ "query": "Taiwan AI startups", "max_results": 5 }

POST /fetch — Fetch specific page content

{ "url": "https://example.com/article" }

本地 Bing 搜尋服務

本地 Bing 搜尋服務，透過 Playwright 抓取搜尋結果並以本地 LLM 摘要回答問題。支援 MCP Server 與 HTTP API 兩種接入方式。

需求

• Python 3.10+
• Microsoft Edge 或 Google Chrome
• 本地 LLM 服務（LM Studio、Ollama 等，需相容 OpenAI API 格式）

安裝

pip install fastapi uvicorn playwright beautifulsoup4 openai python-dotenv pydantic "mcp==1.6.0" "starlette>=0.40.0,<0.48.0"
playwright install

mcp >= 1.7 會與 fastapi 產生 starlette 版本衝突，請固定使用 mcp==1.6.0。

設定

複製 .env.example 為 .env 並填入設定：

LLM_BASE_URL=http://localhost:1234/v1
LLM_API_KEY=lm-studio
LLM_MODEL=your-model-name

BROWSER_CHANNEL=msedge   # msedge 或 chrome
HEADLESS=false

HOST=127.0.0.1
PORT=61500

啟動

MCP 模式（Claude Desktop、Cursor、Windsurf）

python main.py mcp

在 claude_desktop_config.json 加入：

{
  "mcpServers": {
    "bing-search": {
      "command": "python",
      "args": ["C:/path/to/main.py", "mcp"]
    }
  }
}

HTTP 模式（LangChain、AutoGen、OpenClaw）

python main.py http

啟動後：

• API 文件：http://127.0.0.1:61500/docs
• Tool Manifest：http://127.0.0.1:61500/.well-known/ai-plugin.json

API

POST /search — 搜尋並摘要

{ "query": "台灣 AI 新創", "max_results": 5 }

POST /fetch — 抓取指定頁面內容

{ "url": "https://example.com/article" }

License

Apache-2.0